Quad Core Chips From Intel and AMD

lubricated writes "According to the San Fransisco Chronicle, in an effort to one-up AMD, Intel will be coming out with 4 core cpu's in 2007." From the article: "Chips with two cores have been the latest rage, with both Intel and AMD selling those microprocessors as their high-end offering. Apple Computer Inc.'s new iMac, which started selling last month, uses the dual-core chip ... Not to be outdone, Randy Allen, AMD's corporate vice president of server and workstation division, said Friday that his firm is working its own quad-core processor for release next year."

Comments

The new race

[4D6963]

Say bye to the race to the Gigahertz. Say hello to the race to the core count

Re:The new race

[Anonymous Coward]

Say goodbye to insightful commentary on /.

Say hello to retarded assclown karma whores.

Re:The new race

[Jeff DeMaagd]

Say bye to the race to the Gigahertz. Say hello to the race to the core count

Really. It does seem that there's only so much that can be done to increase the clock. I hope this gives an impetus to improve multi-CPU software performance.

Re:The new race

[4D6963]

"I hope this gives an impetus to improve multi-CPU software performance."

Yes, but there is the problem. With the Gigahertz race, you were sure to be able to enjoy the benefit. With multiple cores, you need software able to use these cores, am I wrong (I'm not really sure of what I'm talking about)? And so far we can't always fully exploit these multiple cores, am I wrong?

Re:The new race

[Wyatt Earp]

Not sure that with gigahertz we got to enjoy the benefits. Word on my 233MHz G3 worked as well as it did on 800MHz G4 as it does on my 2x2GHz G5 and as well as it does on my 3.2GHz P4.

I really think in the "megahertz" race we didn't really enjoy the benefits in all areas of software. vi, emacs, text editor x don't really benifit from 3GHz over 333MHz. Someone who just pops open Word or Word Perfect and an email client doesn't benifit from something zoom zoom high GHz.

On that note, quite a few things on OS X work better for CPU/usage on a pair of slower CPUs than on one fast CPU.

Re:The new race

[Anonymous Coward]

My word, people! When will it end? HOW will it end? can't you see these multiple cores are tearing slashdot apart?

Re:The new race

[4D6963]

"Word on my 233MHz G3 worked as well as it did on 800MHz G4 as it does on my 2x2GHz G5 and as well as it does on my 3.2GHz P4."

What you just said reminds me of something I read before, it was something like "Computers in 93 used to run Word as fast as on "I really think in the "megahertz" race we didn't really enjoy the benefits in all areas of software. vi, emacs, text editor x don't really benifit from 3GHz over 333MHz."

Suuure. It didn't benefit much for NetHack neither, but what kind of improvement can you expect from vi that is not mainly design decisions or stuff like that?

vi now can only get as good as it could have got ten years ago because it meets more than the requirements it needs to do anything it can do. But many areas of software can't say they met more than they need, and that's obviously why computers need to go faster for some type of applications to improve.

I fail to see your point anyways, is it that computers don't need to go faster, just because we can run vi or Word fast? Are that type of applications all you need, or do you also use more needy applications like for example a web browser that would have runned more painfully on your 233 Mhz G3?

Re:The new race

[lubricated]

This has many interesting implications in software development.
As someone that designs algorithms this is something I look forward to with much excitement.
I for one welcome our new kilo-core overlords.

Re:The new race

[4D6963]

What.. are.. you.. talking.. about?

Re:The new race

[thewiz]

Don't you mean Core Wars?

Re:The new race

[miffo.swe]

Im selling my stove and getting one of theese ASAP!

Seriously, how much power will they consume?

Re:The new race

[luvirini]

Even if you did not need a fast computer you were able to enjoy gigahertz race, just buy the slowest computer there is, and enjoy the other side of it.. slower things became cheaper.

Re:The new race

[WuphonsReach]

There are plenty of applications that can exploit multiple CPUs or cores. Most video encoding packages will encode a single stream using as many cores as possible (encoding is easily done parallel).

Other apps get an entire core to themselves, while the O/S and background tasks run on the other cores. This makes for a snappier system because the UI threads aren't competing for CPU time on a clogged CPU.

Re:The new race

[Wyatt Earp]

My point is, that the parent I replied to said all software benifits from more MHz, that's not true since not all software does benefit, somthings run just fine on slower machines.

Re:The new race

[4D6963]

I wasn't talking about the software benefits from more Mhz. I was talking about benefiting, understand, being able to use, those new cores, compared to being able to use the new MHz, back then.

The point was that, if your computer had twice as much MHz, basically, you could make let's say twice as many multiplications in as many time. But if you have a quad-core CPU, given the same amount of GHz, you will only be able to make 4 time as many multiplications if your software is written for using those 4 cores, otherwise, you're fucked.

Nothing to do with whether in your everyday use you benefit from more power or not.

Re:The new race

[Doctor Memory]

With multiple cores, you need software able to use these cores,

Not necessarily, which is to say, not at all. If all you use your computer for is one CPU-intensive thing (games or rendering or Mathematica), then yes you won't get all the benefit. But computers today run so many tasks simultaneously (one per window, plus init, swapper, inetd, cron, lockd, statd, X server, plus whatever else you're doing) that you'll get some benefit regardless what you run. I still have issues occasionally when my XP box at work will freeze when the network gets loaded. With a true multi-core (none of this HT stuff) or multi-CPU box, you'll at least be able to fire up a shell (or the Task Manager) and find out why something's so slow.

Re:The new race

[4D6963]

So I got one question that I've been wondering since the Dual G4's appeared, why does it seem important to dedicate a core to a task and another to another over making run both on one core the way we know it (with interrupts so each gets its fair share of time)? Of course the sum of the power you can get to with let's say two cores is surely more than what you can get with only one, but is this the only reason?

Or is there some reason that would make having two 1 GHz cores better than one 2 Ghz core?

Re:The new race

[4D6963]

"This makes for a snappier system because the UI threads aren't competing for CPU time on a clogged CPU."

Yeah, I imagine it to be a nice advantage, but it makes think of two things. Firstly, why not use all the other cores instead of one? And then, if you only need one core on a multi-core CPU, why won't you need a limited percentage of a one-core CPU for each thread so it acts the same?

Maybe what I mean is not clear, so imagine a quad-core CPU. Your threads can use each core etc, etc.. On a single-core, why don't you "cut" your CPU into 4 "virtual cores" so for example one thread that would act like it's only using one "virtual core" would actually be using 25% of the CPU time slices?

I'm not proposing realizing virtual cores tho, it would come back to having the disadvantage of having multiple cores and the power of only one CPU, but as a purely theorical question i'd like to know what it changes (aside from the fact that you'd need a powerful CPU to get the same result with 25% of its time slices as with one core of a quad-core CPU).

Re:The new race

[Wyatt Earp]

If the extra speed, extra cores don't do anything make the end product better for the user, in whatever the user is doing, then there isn't a point to it.

That's one of the problems with hardware/software right now, most of the software doesn't get any benefit from the speed/cores, yet the hardware makers are shoving more speed/cores down the end user's throat and keeping the price of computing higher.

Re:The new race

[hobbes75]

>Or is there some reason that would make having two 1 GHz cores better than one 2 Ghz core?
Mainly power consumption (and therefore heat) that increases super-linear with the frequency.

For most single tasks, a single core with higher GHz count is better for finishing it in shorter real world time. There are some optimization problems that profit from a super-linear speed-up with processor count but these are the exception.
If there are several processes running on the system at a load > 1 (and full CPU usage, not I/O waiting), an additional core reduces the significant task switching overhead.

Re:The new race

[damiam]

If you hadn't noticed, the price of computers has decreased over recent years. No one is shoving cores down your throat.

Re:The new race

[Millenniumman]

I can definitely see credit card sized computers with 10^(10^100) core nanoprocessors. By then, computer enthusiasts will all know that the number of cores is for marketing, and what really matters is yottahertz.

Re:The new race

[Anonymous Coward]

"Say bye to the race to the Gigahertz. Say hello to the race to the core count."

I really dont think more cores are going to make performance better, how is it going to make performance better if you have 4 cores and no increased memory bandwidth? It doesn't add up.

Re:The new race

[timeOday]

With multiple cores, you need software able to use these cores, am I wrong

The transition from single-threaded to multi-threaded is fundamental, and will require a permanant increase in code complexity that we'll all have to learn to live with. However, the transition from 2 to 4 to more should be little or no trouble. At this point only a foolish programmer would think in terms of exactly 2 cores instead of N cores.

The main mistake I think people are making is the idea of having each thread do something different, e.g. one thread for graphics and one for AI. To harness a large number of cores equally, we need libaries which divide up big repetitive tasks (say, collision detection or matrix multiplication) into a large number of chunks. Of course you can't write heavily procedural logic that way, (say, a word processor), but for the most part that stuff runs fast enough on one core anyways.

Re:The new race

[4D6963]

"most of the software doesn't get any benefit from the speed/cores"

Ever heard about video games? yet it's just an example

Re:The new race

[4D6963]

"an additional core reduces the significant task switching overhead."

Does switching from a task to another sums up to a significant percentage of CPU time slices?

Re:The new race

[4D6963]

"you'll at least be able to fire up a shell (or the Task Manager) and find out why something's so slow."

That's why I been wondering for years, why not just prevent something from using 100% of the CPU? I mean, if we let a process only take up to let's say 99% of the CPU, we'd still have a system reactive enough to get to the task manager and eventually put and end to the process that would otherwise make anything else so slow that you'd have to reboot. It's not even about priority, just about preventing a process from using 100%, and depriving it from a non-significant processing time, but that would be significiant for the other tasks.

Re:The new race

[shufler]

On a single-core, why don't you "cut" your CPU into 4 "virtual cores"

This is the same as dividing the speed of yor processor by four. Splitting a single processor into "virtual" time slices isn't the same as having mutliple cores, as a single instuction is still only being executed at any single moment in time. Your example would result in a scenario where no process can use more than 25% of the processor. Multiple cores--like multiple processors--operate independantly of each other. This means four instructions are run simultaneously. If you run an application that uses 100% of only one of the cores, your computer will not hang as there are other cores it can use.

Re:The new race

[4D6963]

"The main mistake I think people are making is the idea of having each thread do something different"

Totally agree. I once read a comment by a guy who was talking about a 8-core gaming system, and suggesting that one would be dedicated to graphics, one to the AI, one to sound, one to controlling the controllers, lol, some stupid crap like that. If you're gonna use a whole core just for controlling a keyboard or a gamepad, needless to say that you're a fool.

Of course they need to be put to contribution in a way that all the available power is used when needed. Which makes me think, let's say you run a CPU hungry application, couldn't by some way the OS cut the work to be done between each core so it would run concurrently on many cores without the application program to have to modify anything, just having the OS to deal with it?

I realize that the main problem with my question is, if your application is doing serial processing (for example basing it's next computation on the result of the previous one) it's not gonna be easy to make it work concurrently on several cores, but to a more or less small extent, couldn't something be done about the OS so it could make some applications written for a single core take advantage of multiple cores?

Re:The new race

[Anonymous Coward]

I work for a company producing chips containing eight fast MIPS64 CPUs which support 32 threads. These 8-core chips are running 32-way Linux right now in customer systems. They have been on the market since May 2005. You could put a 64-core supercomputer on a board now, although it would not fit in a PC case. Intel and AMD are actually playing catch-up.

Re:The new race

[4D6963]

I know all that, that's why I said "aside from the fact that you'd need a powerful CPU to get the same result with 25% of its time slices as with one core of a quad-core CPU".

"If you run an application that uses 100% of only one of the cores, your computer will not hang as there are other cores it can use."

If you run an application that uses 100% of only one "virtual core", your computer will not hang either, that's the point of my previous comment. OK, there you're only using 25% of the power of your single-core CPU, but after all, when using 100% of a core out of four-cores, you're still using 25% of the power of your CPU.

So why is one acceptable and not the other? For scale reasons? (understand, using one core of a recent four core CPU gives u a good performance, as using 25% of a recent single core CPU gives you a poor performance, so the difference is in a way only a matter of scale)

Re:The new race

[Anonymous Coward]

Well let's get the jokes out of the way here..

640 cores ought to be enough for anyone.

I, for one, welcome our new 64 core overlords.

in soviet russia, multiple cores use you!

in korea, only the old use multi cores.

so how many fusion reactors does that new quad core system need to post?

great now i can cook my eggs 4x faster!

Re:The new race

[Anonymous Coward]

YHBT. YHL. HAND.

Re:The new race

[Anonymous Coward]

FWIW technically speaking, the most 'cores' a video encoder can sucessuflly use when converting a 640x480 resolutions stream into mpeg-4 format is 1200. however, the RAM requirements on a 1200 thread video encoder become somthing from a nightmare, it could be as high as 64 GB of ram in a poorly designed encoder. of course, if your system has 1200 cores, not having 128 GB of ram would make you look cheap and pathetic.

in large parts this is why 'multi core design' is a 'path of deminishing returns' it's been pretty well shown that after 16 cores/processors there isn't much room for improvement except in very highly specialized OSes and applications. quad core in a 4 socket board offer all the cores that most applications oses etc would have a hard time figuring out a way to get better performance from 'more' cores.

Re:The new race

[imgod2u]

It's a lot easier to make 4 cores into a 2 GHz chip and have one process run on one core, than to make an 8 GHz chip and let software take up 25% of that processor's time. It's both hard in terms of making that 8 GHz chip and also of getting software vendors and the OS to agree on allocation. Also, some software might want more than that 25% timeslice. And the user, being in agreement, would want that as well. It is then a simple matter of a register setting to let that process use more on the 8 GHz chip, not so with the multi-core chip.

Re:The new race

[4D6963]

Yeah, of course. I was obviously not suggesting that to be actually done. I was just wondering how comparable to multi-core CPU's such a design decision would be.

Re:The new race

[lubricated]

>> Or is there some reason that would make having two 1 GHz cores better than one 2 Ghz core?

I would generally take a 2Ghz core. Not much slower for two tasks, twice as fast at one. The problem is that cpu makers have stopped being able to make cpu's run faster, it's plataued(sp?). More cores will require new software techniques. In theory, you could program certain algorithms so that they increased linearly with each core(not super-linearly as another poster suggested, that doesn't make sense). With good software techniques you could be infact increasing your speed so much that 2x1Ghz = 2Ghz, but that's in the future, In a practical present sense, less switching and having only one cpu doing IO, means a more responsive computer.

Re:The new race

[hobbes75]

"Does switching from a task to another sums up to a significant percentage of CPU time slices?"

No, the overhead is typically very small, below or around 1%. It depends on OS, the time slice frequency and the kind of context switch that has to be done. Context switches between threads of a single process are very cheap, switches between processes are much more expensive because the MMU has to be adjusted for the new process.
With more cores, the time slice frequency can be reduced to achieve the same amount of "perceived parallelism". E.g. reducing the frequency from 1000Hz to 500Hz cuts the overhead by factor of 2.
Reducing the time slicing frequency can speed up CPU bound processes e.g. number-crunching applications. The gain is normally small.

Re:The new race

[calidoscope]

Maybe what I mean is not clear, so imagine a quad-core CPU. Your threads can use each core etc, etc.. On a single-core, why don't you "cut" your CPU into 4 "virtual cores" so for example one thread that would act like it's only using one "virtual core" would actually be using 25% of the CPU time slices?

The rational for multi-core CPU's is that 4 slow cores can do the same amount of work as a fast single core system and consume less power. A related rationale is that you can't get a single core system that's 4X faster per core than a 4 core system. With those two exceptions, there is nothing to prevent what you're proposing.

The CDC-6000 series peripheral processor had an interesting twist on the virtual core - the ALU was time-sliced amongst 10 register sets so that it appeared to be 10 processors. The Sun Niagara does something similar - each of the 8 cores has 4 register sets, allowing for very rapid switching between threads. Recently saw a usenet posting stating that a single T2000 performed twice as fast as a dual Xeon box - and the Niagara CPU uses less power than a single Xeon.

Re:The new race

[Anonymous Coward]

I call bullshit.

You are lying through your teeth.

My wife uses Word (and I'm not talking for 1 page letters) and she upgraded her 400 MHz G3 to a 433MHz G4 and eventually just ran up against her computer being too damn slow. 2 orders of magnitude kind of slow. I had to run out and get a MacMini when she was hitting a deadline and it made all the difference. Now she is running the fastest portable Apple has ever shipped.

Claiming Word runs the as well on a 233MHz G3 as it does on a 2x2GHz G5 is so out-of-bounds I can't believe you are delusional or even just typing up 1 page letters with no graphics and 1 font and somehow you think that makes you qualified to speak of Word's speed on different hardware. You are lying.

Re:The new race

[babble123]

The main mistake I think people are making is the idea of having each thread do something different, e.g. one thread for graphics and one for AI. To harness a large number of cores equally, we need libaries which divide up big repetitive tasks (say, collision detection or matrix multiplication) into a large number of chunks.

In the jargon, you're talking about the difference between task parallelism and data parallelism.

Re:The new race

[Minwee]

So which company will be the first to introduce a giga-core chip?

Re:The new race

[Minwee]

"Of course you can't write heavily procedural logic that way, (say, a word processor), but for the most part that stuff runs fast enough on one core anyways."

Even if your main applications are single-threaded and can only use one CPU, that still leaves you with three cores to run nothing but Norton Anti-Virus.

Some people will notice a slight increase in performance as a result.

Re:The new race

[fvongraf]

Who cares how many cores they have when the Northbridge is a bottleneck with one core. I'd like to see improved chipsets and faster memory & I/O!

Re:The new race

[Anonymous Coward]

smack your wife around the face and teach her to use vim.

Re:The new race

[Sneftel]

Word...vi...emacs...text editor x...Word Perfect...
My goodness, you're right. It's almost as though the benefits of faster processors are primarily realized by software which is processing intensive! But... but that cannot BE!

Re:The new race

[drsmithy]

That's why I been wondering for years, why not just prevent something from using 100% of the CPU?

Because almost all the time, you *want* a process to be using 100% of the CPU.

I mean, if we let a process only take up to let's say 99% of the CPU, we'd still have a system reactive enough to get to the task manager and eventually put and end to the process that would otherwise make anything else so slow that you'd have to reboot. It's not even about priority, just about preventing a process from using 100%, and depriving it from a non-significant processing time, but that would be significiant for the other tasks.

That's exactly what priorities are for.

Re:The new race

[glitch23]

Word on my 233MHz G3 worked as well as it did on 800MHz G4 as it does on my 2x2GHz G5 and as well as it does on my 3.2GHz P4.

You must be one of the rebels that MS targets for extinction. You aren't supposed to use the same version of Word on all of those CPUs. You are supposed to upgrade to the newest and slowest version so your CPU will have something to do while your text-based email is being transferred to your mail server.

Re:The new race

[saneman]

Ever heard about video games? yet it's just an example

Hz so good for encoding and decoding of video files too.

Re:The new race

[Luminous Coward]

Not sure that with gigahertz we got to enjoy the benefits. Word on my 233MHz G3 worked as well as it did on 800MHz G4 as it does on my 2x2GHz G5 and as well as it does on my 3.2GHz P4.

Faster processors do help in several areas, e.g. video and audio codecs. Have you tried H.264 encoding on your trusty G3?

Re:The new race

[jadavis]

couldn't something be done about the OS so it could make some applications written for a single core take advantage of multiple cores?

That would seem to be VERY difficult to do without introducing many strange bugs. Software is made up of text, data, stack, and heap segments, all in a virtual address space. For a single-threaded application, there is one instruction pointer that points to the next instruction to be executed. So, the whole thing is very linear.

However, there are potentially things that a compiler could do. It has a lot more information about what variables depend on what other variables. Still hard, but seems more managable at the compiler level.

Re:The new race

[4D6963]

"That's exactly what priorities are for"

Doesn't always work out good. Haven't you ever be stuck with a process with normal priority but that makes your system stuck anyways? Used to happen to me alot when previewing partial videos from eMule with VideoLan, even when I set VLC's priority to Low in the taskmgr. Could take me about 5 minutes to get to the taskmanager and to kill the process

Re:The new race

[4D6963]

Indeed! Or even computing alot of big convolutions, or computing those distributed computing projects, or emulators (and with a 2 GHz Athlon I still can only run Red Baron from 1983 in MAME at 30% of its normal speed), or even running Vista and its hungry GUI.

Examples are countless, but if you only need to run Word under Windows 95, then alright, maybe you don't need a 3.6 GHz Pentium

Re:The new race

[heinousjay]

What is the difference between what you're saying and an OS scheduler?

Re:The new race

[Anonymous Coward]

runned?

Fuck, American education has completely given up.

Re:The new race

[Sique]

First: It is nearly impossible to automatically convert a single thread application into a multithreading one. Multithreading requires a lot of things to be guaranteed that aren't in a single thread application. One of the most important is dealing with reentrance: What happends if two threads are running the same code at nearly the same time and try to access the same global variable? Which goes first? Which writes the result first, which last? What happens if the first thread later wants to read from the global variable again and the second thread has changed the contents? Is it possible to have the global variable twice, one for each thread? Or would that destroy the semantics, and in the end you are having just the same program running twice (which is quite different from having the program run in parallel on different cores, speeding up the execution twofold).
Maybe a compiler can spot complete data independence between two different parts of the program and put them in different threads. Also the compiler could be part of the operating system as a type of on-the-fly-compiler, compiling machine code into itself (or into micro ops for the special processor, like the Transmeta processor does in hardware) and again spot data independence. But there is no guarantee this will really speed up the program execution. What if the newly found threads are never used at the same time? Then they will be still executed serially.
Imagine the single threaded program as a single rail track. You can't just increase the transporting capacity by having two trains running on the single track. You have to build new tracks to let the trains cross each other. And you have to design a schedule to take advantage of the different trains arriving at about the same time at the crossing point.
But there is hope: In a multi tasking environment like all modern operating systems there is nearly never only a single task running at a given time. A single threaded program would profit from a multi core machine that it will probably run exclusively on a single core, where all other task are sharing the other cores, if it is demanding enough (that means: if it runs with high enough priority to go always first if the OS has to decide which task now gets the next free core).

Re:The new race

[Anonymous Coward]

Not true, because no series of instructions is infinitely parallelizable by introducing more cores, but it can be sped up infinetly by making the cores faster. So to advance in speed, Mhz of the cores will remain important.

Re:The new race

[Anpheus]

You're right, he could spend a fifth, a tenth, or a twentieth of what some people do on PCs, and get a $200 box that can run Word under Windows XP, no less. He could find a PC that could run Linux and OpenOffice for even less.

Re:The new race

[SillyNickName4me]

It could be implemented as an OS scheduler with 4 run queues that each get used in turn.

A somewhat similar concept (virtual cpus) is used in products like vmware tho for somewhat different reasons, and there the actual scheduling is done (for as far as I know) by the OS scheduler.

Re:The new race

[NutscrapeSucks]

Your main problem is OSX, which has a very complex display system, which simply means there's a ton of overhead. I've got an old G3, and Safari can't even keep up with my typing for simple textareas. Much less a complex word processing program like Word (which admitted is also bloated and slow).

Anyway, Word runs fine on a similar 400Mhz PII/Windows system. The worst thing about older PCs is the slow harddrives, not the CPUs. Windows' much simplier display model still works fine on those boxes.

(And IIRC, G3s and G4s were identical for integer work, so your "upgrade" really didn't do anything for Word.)

Re:The new race

[pitc]

I beg to differ... explicitly taking advantage of multiple CPUs is something that should be avoided for the vast majority of user space apps. Put the code to handle multiple CPUs into the kernel, and maybe a few shared libraries, and everybody wins.

Re:The new race

[Chandon Seldon]

The big difference is that a system like that would be much more predictable than a true mult-core setup because you never have two threads running at once. You wouldn't have any of the race condtions of a true multi-core setup - in fact it would be just like a current single core machine if each process was throttled to no more than 1/4th of the time slices.

Re:The new race

[4D6963]

The difference well.. it's mainly that I never heard of an OS scheduler before. I wouldn't spit on an explanation of what it is :-)

Re:The new race

[4D6963]

Actually he could get himself a $5 Mac Plus with Mac OS 6.0.8 and ClarisWorks 3. Some people even give away their Mac Pluses for free.

Re:The new race

[hevenor]

As far as writing for multiple cores are concerned...you generate the most parallelism by separating all code that works on mutually independent segments of data and not by thinking in terms of how many processors you have.

Most classically trained Computer Scientists are at least taught to develop in this way and I think that most professional software is developed using multiple threads as it is. Even on a single CPU machine a thread gets a window in which to execute and then is suspended to let another run. The only difference when you add more cores is the thread scheduler.

You are right though, to see a performance boost from one specific application run in isolation it will need to be written using threads. Luckily our operating systems these days are multi-threaded so having more cores can only help :)

Re:The new race

[heinousjay]

This is an extremely simplified explanation, and is a total lie in reality. However, it is a useful lie, and a good place to start.

Essentially, a multitasking operating system divides the processing time of the computer amongst several processes. It does this by letting the process run for some deterministic amount of time, and then pausing that process and switching to another one. Each process labors under the illusion that it has the whole machine to itself. Maintaining this illusion is one of the primary purposes of the OS.

A process is basically analogous to an application, although that's not 100% true.

More information is available, from people who spent more time explaining things. Google has more, naturally. This topic is as deep as you feel like swimming.

Re:The new race

[astralbat]

Your right, but your missing something. In the context of a multi-tasking operating system, applications are constantly switching in and out of the processor. If you had 4 cores, you can be running 4 applications simultaneously. Of course, if your running only one intensive single threaded application, then your unlikely to see the difference.

Also, I believe that multi-threaded applications will come of age as developers realize there is a greater CPU resource available if they can divide the work load up. Gaming for example is one area where this will be extremely important, but there are certain areas such as tight AI algorithms where dividing work will be difficult.

Re:The new race

[jadavis]

Not much slower for two tasks, twice as fast at one.

Not true at all. Most CPU-bound applications are actually memory-bound. CPUs spend most of their time idle, waiting for memory. Idle at 2GHz is no better than idle at 1GHz.

Having 2 cores means you can parallelize independent tasks. 2 1GHz cores, in some cases, could be almost twice as fast as one 2 GHz core.

Not only that, but there are a lot of other factors, like how big the pipeline is and how it's constructed.

Re:The new race

[lubricated]

>> Most CPU-bound applications are actually memory-bound.

Then they would be called memory bound applications. In my line of work applications are strictly cpu bound.

Re:The new race

[jadavis]

Exactly. I was responding to someone who thought that the OS could do something to automagically multi-thread. I was trying to say that once you have machine code that can access any memory location, it's pretty much impossible to try to multi-thread it then.

However, it's a little more possible at the compiler level in, e.g., a java program. Let's say you have some code like:

[...]
x = factorial(100);
y = fib(112);
return x + y;
[...]

If both factorial() and fib() have no side effects, the java compiler could run them in separate threads and join the threads right before the return statement. The compiler could guarantee that those functions have no side effects because in java, you can't dereference an arbitrary memory address. Perhaps a C compiler could guarantee that, but that would be a lot more challenging and bug-prone.

Re:The new race

[jadavis]

Ok, fair enough, but you should have probably mentioned that, since it's relevant to the post you responded to, and not the typical situation.

And 2GHz still isn't twice as fast as 1GHz, unless maybe if you fit all of your data in L1 cache, and all other processor properties are identical.

Re:The new race

[StikyPad]

I'm with you. I'm loathe to spend more money on a multicore CPU until I see ubiquitous multithreaded programming. Maybe (a lot of people) doing that will have the unintended consequence of slowing or killing the multicore market altogether, but they should have thought of that when they made the market. There should be more incentive to purchase a product than "this will be highly useful at some point in the future when programs take advantage of it, at which time you'll need to buy a new one anyway."

Re:The new race

[ChrisGilliard]

This is all correct. The perfect "N-core" application is a webserver. That's why I think this multicore stuff really lends it self to the server market more than than the desktop market. The desktop market is being eaten at both ends. One end is the laptop market, the other is the server market. I'd imagine that in 10 years, most computers that consumers buy will resemble laptops more than desktops. These laptops will be measured on other metrics, like battery life, graphics display capabilities, user input devices, etc. The servers that the laptops connect to will be where most of the actual processing takes place.

Re:The new race

[Doctor Memory]

I remember VMS had a decent approach to that. Something like, if your process was idle for more than a few seconds, it would temporarily raise the priority (after all, you're not doing anything, so it's free). Then, once your process woke up, it would degrade the priority back to its assigned level. All normal tasks ran at priority 4, but if you were in the editor and not typing, your priority might get bumped up to a six. Someone else's compile might be soaking up 100% of the CPU, but as soon as you hit a key, you got priority. By the time you'd entered half a line or so, your priority was back down to 4 and you were slogging it out with the compiler job, but initially you got pretty much instant response, which told you that the machine wasn't wedged.

Re:The new race

[4D6963]

Souns good, but as far as I read M$'s doc about it, Windows does pretty much that to, about boosting priorities and all that. I guess it just doesn't handle it as nicely as the example you cited.

Re:The new race

[cthulhu11]

You've never twiddled thumbs waiting for OpenOffice to fire up, I'll bet.

Re:The new race

[default luser]

Well, look at it this way: early superscalar x86 processors went through the same issues. When parallell execution pipes were added, the effects weren't really noticed overnight.

The original Pentium was a superscalar, in-order core with two integer pipelines. To put it simply, the design was like two 486 integer pipelines slapped together. But the real-world performance of the Pentium was about the same as the 486 when it came out, because the Pentium's pipeline scheduler was simple and picky. The scheduler would only let two instructions down the pipe if:

* Both were simple instructions
* No read-after-write or write-after-read register/flag dependencies existed between them.
* Neither can had a displacement and an immediate
* Instructions with prefixes can only occurr in the U-pipe
          (except for branches on condition jz,jc, etc.etc.)

Basically, it was a pain in the ass to get more than one integer instrution per clock out of the Pentium. Eventually, compilers caught up, and the Pentium's performance improved. Also, hardware caught up, and the Pentium Pro introduced out-of-order execution with a robust scheduler to handle these issues even in unoptimized code.

The multi-core situation actually has a better initial handle on the situation than superscalar processors did when they were released. Modern operating systems are already optimized for multiple processing threads, and lots of professional applications have been multithreaded for years, so right out the door we have solid multitasking performance advantages over single-core processors. Even games have begun to catch up, with the release of multithreaded titles like Quake 4 and Serious Sam 2.

But there is a limit. Unfortunately, most "multi-processor" workstations have been dual-processor workstations, and thus most applications are optimized for two threads only. So, unless you go nuts with the multitasking, quad cores are going to hit a perfoemance wall. And unlike superscalar processors, which simply required a smart compiler, optimizing programs for lots of execution threads is going to be time-consuming for almost every application.

Re:The new race

[JacobO]

I have to ask where you got your numbers from. Why 1200? Why only 16 cores/processors?

How specialized is it, would you say, to run Oracle? I've seen Oracle scale out quite nicely to accommodate more users on a 64-way Sun box.

I suppose that if you're talking about PC hardware, or PC-derived hardware such as Intel server mobos for Xeon processors, then I guess you might be right about the limitations. Other architectures have this problem largely solved.

If only...

[Hextreme]

These new X-core chips could use less power than their predecessors. Everything is so fast these days, that's almost a larger concern than the speed or cache count. "How hot does it get and how much power does it consume?" is the real measure of success here.

Re:If only...

[Anonymous Coward]

These chips can offer a great deal in concern to using less power.

1) Only using the cores that are neccesary
2) Intelligent cache sharing, such as that used in the Core Duo (big cahce shared will mean less die-estate used for cache which reduces heat produced)
3) More cores means that each core can be lowered-clocked in many situations and maintain similar performance as a much higher clocked single-core system. The higher clocked system will use more energy due to higher current leakage at higher frequencies.
4) I have heard that processors use less power when they are cooler (same load, different operating temp) - so having multiple cores and cycling the workload to a cooler core when the current one reaches a non-optimal temp would save power when all the core are not being utilized

As technology increases I imagine the wattage of a single core and mutlicore processors will converge. This is will be due to systems better saving energy when parts of the processor are not in use thus negating the negative effects of having more tranistors on a multi-core chip.

Re:If only...

[Ravenrage]

granted those are valid concerns but why can't they improve the other things(processing power and such)? i mean granted i don't want a furnace inside my computer but multi-cores would be helpful think about it 1 cores for video encoding 1 for audio and the others for the os and it's tasks and still be able to watch a movie with out skipping. sorry to burst your bubble but not everyone feels the same way as you.

Multi-cores

[acslat3r]

I am looking forward to multi-core systems. I have an athlon 64 dual core 3800 using windows for my main ebay computer and it can pretty much handle anything i throw at it. It will be interesting to see how the motherboards of the future look and how the memory is allocated since I would assume all of these cores sharing the same memory has to have more of a performance penalty. Adobe premiere recognizes the dual core during startup but I don't know of many programs that use both cores..i guess it just splits the load between them. I would assume multi-cored processors will sharply scale up in price due to the lower yield rate from effectively making two-four-eight processors at once on a single die.

Re:Multi-cores

[keyrat rafa]

I have an athlon 64 dual core 3800 using windows for my main ebay computer and it can pretty much handle anything i throw at it.

wtf is an ebay computer and why would it need a fast processor?

Re:Multi-cores

[pwroberts]

I have an athlon 64 dual core 3800 using windows for my main ebay computer and it can pretty much handle anything i throw at it.

wtf is an ebay computer and why would it need a fast processor?

It was subliminal messaging. You are now ~12% more inclined to hunt for amazing last minute bargains at ebay.com! And so am I.

Re:Multi-cores

[Compholio]

... but I don't know of many programs that use both cores..i guess it just splits the load between them.

Under Windows it will only use one of the cores for most operations (a small number of programs can use both). Linux SMP kernels will split up the tasks for the two processors as best as it can, supposedly Solaris and the BSDs will do it too but I've never seen those guys at work.

Re:Multi-cores

[Dashing Leech]

"It was subliminal messaging."

Oh that's a load of rich, creamery butter.

Re:Multi-cores

[Bing Tsher E]

My eBay computer is the one that runs sniping apps. I don't know why processor speed would be an issue, unless you have a LOT of bandwidth to spare and a LOT of items to snipe, though.

Re:Multi-cores

[quanticle]

/*I would assume all of these cores sharing the same memory has to have more of a performance penalty.*/

Depends on your architecture, really.  Intel chips might be hurt because of their sharing an external memory controller.  AMD chips would experience little or no performance penalty, because they have the memory controller on-die, and the chance of two cores accessing the same memory address is small.

/*Adobe premiere recognizes the dual core during startup but I don't know of many programs that use both cores..i guess it just splits the load between them.*/

As sibling posters have indicated, that is not how it works.  Unless an application is specifically written to be multithreaded, it will only use a single core.  Multiple cores allow you to have multiple single-threaded applications open without as much of a performance hit, but multi-core will not significantly improve the performance of a given application, unless the app is rewritten/recompiled to take advantage of multiple cores.

Re:Multi-cores

[Anonymous Coward]

Yeah, and why would he be "throw[ing things] at it"?

Re:Multi-cores

[Anonymous Coward]

I also have a 3800+ dual core and the only program that pegs cpu usage to 100% is dvdsanta...decodes divx to dvd at the rate of 0.6 seconds per second of movie...

Re:Multi-cores

[jordie]

Unfortunately it worked, I'm now browsing ebay in another tab.

Re:Multi-cores

[Daniel Rutter]

Either
(a) he's running a truly awesome amount of auction management and/or sniping software for a whole bunch of auctions, or
(b) he's running eBay.

Re:Multi-cores

[Anonymous Coward]

wtf is an ebay computer and why would it need a fast processor?

it's for the smooth-scrolling, of course!

Re:Multi-cores

[McFadden]

Unfortunately it worked, I'm now browsing ebay in another tab. Dammit. I'm now pining for a clear cola drink. Clever stuff this subliminal advertising.

Having multiple cores ...

[Gaima]

... inside a chip is like having more than one engine under the hood of a car.

Except it's quite useless with front wheel drive.

Re:Having multiple cores ...

[TubeSteak]

... inside a chip is like having more than one engine under the hood of a car.

Except it's quite useless with front wheel drive.

Except when you're driving in the snow.

What I really meant to say is WTF kind of analogy is that?
FWD, like RWD has its pros/cons and each has its place.

Do you even know why FWD was such a god-send back when it was first introduced? It was easier to manufacture & install in cars, it provided better handling characteristics for the vast majority of drivers (oversteer will fuck up your day in a heartbeat, understeer is more 'predictable'), you got better fuel efficiency due to a shorter drivetrain, etc etc etc.

Allow me to recreate your quote, but with even more esoteric comments: ... inside a chip is like having more than one engine under the hood of a car.

Except it's quite useless with a 2-barrel carburetor
Except it's quite useless without a limited slip differential
Except it's quite useless with bias-ply tires.

You get my point yet?

Re:Having multiple cores ...

[ArbitraryConstant]

"(Having multiple cores)... inside a chip is like having more than one engine under the hood of a car."

No, it's like having multiple cylinders inside an engine.

Re:Having multiple cores ...

[damiam]

I don't think the GP was slamming FWD. He/she was was just pointing out that two engines in a FWD car would be pretty much useless, which is quite true. Actually, two engines are useless in any car, unless that car is a Bugatti Veyron.

Re:Having multiple cores ...

[Zemplar]

"No, it's like having multiple cylinders inside an engine."

Let the dead horse rest. That was another totally brain-dead analogy.

Re:Having multiple cores ...

[Akaihiryuu]

My car is front wheel drive and it has two engines. http://www.toyota.com/prius/ Course, one is electric...but still, it disproves that point.

Re:Having multiple cores ...

[Anonymous Coward]

Not to pick nits, but your car has an engine and a motor. The two are related, but I think the term 'engine' implies something above and beyond what a motor does--that being that an engine converts thermal energy directly into mechanical work, whereas a motor creats motion from any source of power, be it mechanical (spring), electricity, or some sort of fluid power (hydraulic motors, water wheels and turbines, for example).

So, in otherwords, an engine is also a motor, but a motor cannot be an engine... But engine has come to be synonymous with motor amongst the general population, it's still technically incorrect.

Re:Having multiple cores ...

[iotaborg]

You've never seen the Jeep Hurricane then, two V8 engines.

Link

Re:Having multiple cores ...

[Saint V Flux]

"it provided better handling characteristics for the vast majority of drivers (oversteer will fuck up your day in a heartbeat, understeer is more 'predictable')," Not really (if you know what you're doing). As an editor in a car mag recently put it -- "Oversteer means that when you go too fast around a corner you'll smash through the fence butt first -- Understeer means that when you go around that same corner too fast you'll smash through the same fence facing forward".

Re:Having multiple cores ...

[Minwee]

That's funny. My car has two motors under the hood, and it uses them quite well. And it's front wheel drive.

Re:Having multiple cores ...

[jawtheshark]

Doesn't the Bugatti have a single engine? Later in the article they say that they merged two V engines into a W configuration. It's still one single engine block.

As for two engines: it is equally useless with RWD.... All cars with two engines that I know of are AWD, which is quite logical:

• Citroën 2CV 4x4 Sahara
• MTM Bimoto TT

Re:Having multiple cores ...

[Gaima]

was just pointing out that two engines in a FWD car would be pretty much useless, which is quite true.

And taking a sly dig at Intel :)

Re:Having multiple cores ...

[Anonymous Coward]

I thought they were referring to a Toyota Prius...

Re:Having multiple cores ...

[gracefool]

That's crap. More cores do not translate into a direct performance increase - there is a lot of overhead, and it only helps at all if the task is parallizable. Also programming for multi-core is MUCH slower/more difficult/bug-prone.

Actually, they were outdone..

[deep44]

Not to be outdone by AMD..? Bad news: AMD made this very same announcement in June/2005, indicating quad-core CPUs would be available 'sometime in 2007'.

Re:Actually, they were outdone..

[Anonymous Coward]

Intel's been talking about multicore since the late 90s. Who cares. Talk is cheap.

Re:Actually, they were outdone..

[Anonymous Coward]

link?

Re:Actually, they were outdone..

[Anonymous Coward]

Google?

Re:Actually, they were outdone..

[Anonymous Coward]

Historicly Intels talk tends to be cheaper.

Re:Actually, they were outdone..

[luvirini]

No.. Intel talk tends to be way more expensive to customers historically...

Re:Actually, they were outdone..

[Anonymous Coward]

Back your shit up with facts and citations.

Re:Actually, they were outdone..

[MojoStan]

AMD made this very same announcement in June/2005, indicating quad-core CPUs would be available 'sometime in 2007'.

But have they demonstrated working quad-core CPUs the way Intel did in TFA? I think an "announcement" is more impressive if they can demonstrate at least a working prototype. From TFA:

To show the product is well on its way, Intel CTO Justin Rattner demonstrated a working server computer with a pair of the new microprocessors, code-named Clovertown.

BTW, Cloverton is based on the Merom/Conroe/Woodcrest architecture, not on the current Netburst/Pentium 4 architecture.

Re:Actually, they were outdone..

[Anonymous Coward]

..and it was (possibly still is) posted on their site that they want eight cores.

I remember drooling a year ago over the very idea of an eight-core, eight-chip Opteron box.

AMD has wiped the eye of Intel by delivering the first useful x86 64-bit chips, the first useful dual-core chips, and I have no doubt that they'll continue to hand Intel their arses. Viva la revolution, at least until AMD becomes a fat, uninnovative monstrosity as Intel has become. But we should be able to start rooting for a new chip manufacturer around that time. ;)

Re:Actually, they were outdone..

[deep44]

But have they demonstrated working quad-core CPUs the way Intel did in TFA? I think an "announcement" is more impressive if they can demonstrate at least a working prototype.

I don't know - that sort of demonstration is completely useless, since there's no way to verify if the demo CPUs are ready for mass production, cool enough for prolonged use, or even whether or not they're faster than dual-core CPUs.

I'm sure there are a few G5 Powerbook "prototypes" laying around at Apple HQ. Clearly, that doesn't translate into a releasable product.

Re:Actually, they were outdone..

[MojoStan]

I don't know - that sort of demonstration is completely useless, since there's no way to verify if the demo CPUs are ready for mass production, cool enough for prolonged use, or even whether or not they're faster than dual-core CPUs.

I think you're right. That reminded me of Intel's demo of their 4.7GHz Pentium 4 in 2002. WTF?

Re:Actually, they were outdone..

[Anonymous Coward]

One of a billion or so: "In a detailed briefing for analysts in New York on Friday, executives at Advanced Micro Devices painted the company as making "irreversible progress" into new architectures, specifically multicore microprocessors and 64-bit processing. Executives confirmed that the company plans to enhance its Opteron enterprise processor line to four cores in 2007, adding focused optimizations to manage power and improve throughput." -- LINK

Good for SGI and Sun.

[CyricZ]

This is a trend that may play out well for SGI and Sun. Both have been building systems which involve a massive number of CPUs for quite a while now. They have the experience that Microsoft doesn't have, for instance.

IRIX and Solaris are known to scale far beyond 4 processors. They're proven technologies that are known to work very well on multiprocessored systems.

SGI could easily use this to their advantage, releasing affordable systems that offer the benefit of IRIX on such machines. If they can come out with a system that appeals to developers and business users, then they could take on Apple, Sun, Dell and others again.

Sun, of course, already offers Opteron-based workstations. A dual CPU entry-level system, with four cores per CPU, could be quite useful. When you factor in the superb quality of Solaris, we could really see some truly fantastic workstations, at a very affordable price.

Re:Good for SGI and Sun.

[Frumious Wombat]

The only question in my mind is whether they have the software engineers working on compilers, autoparallelising tools, etc., to efficiently use these multiple cores. If they don't, then it's the same old "embarrasingly parallel" problem, where your apps don't run any faster, but you can run more of them at once, at the cost of apps competing for cache adn processor bandwidth.

Let's hope this leads to better parallel tools, new algorithms, and new possibilities instead of 'now run 4 simultaneous copies of Minesweeper without a slowdown!'

Re:Good for SGI and Sun.

[cyberjessy]

Are you saying that people just woke up to this "trend"? The industry decided years ago that Mhz war will stop, and everyone will try to put more cores in. Its not like no-one else knew it was coming.

IRIX and Solaris are known to scale far beyond 4 processors.
So does almost every other OS. Linux scales to 1000s of processors, in IBMs supercomputers. Windows 2003 Datacenter supports 64 processors (Which is more than enough for a regular commercial application. In case you want more, instead of scaling up, you should be scaling OUT.) AIX, HP-UX etc also have great support.

If they can come out with a system that appeals to developers and business users, then they could take on Apple, Sun, Dell and others again
SGI competing with APPLE and DELL???? In what segment, but in the figment of your imagination?

SGI?? They lost $100m in 2004, $72m in 2005. They are nearly _dead_ and looking for a sell-out. In many ways they deserve it, I still remember their CD drive being priced 10 times higher than the ones in the market if you wanted to replace one. And of course, being totally proprietary nothing else would fit in. Who is buying IRIX now? And SGI now focuses on Linux.

I don't know who modded you interesting. And I did not know SGI still had fans!

Re:Good for SGI and Sun.

[Anonymous Coward]

You do realize that it isn't REQUIRED that you post to every single story, right? I can only see your post history for today, that is just sad. Please get a life, yours truly...
-- Anonymous Coward

Re:Good for SGI and Sun.

[BigBlockMopar]

And I did not know SGI still had fans!

Surely there's still at least one in the power supply?

Seriously, though, the market being as trend-based and as fast-moving as it is, one can expect that all innovative tech companies will alternate between boom and bust, with the bust times being spent retooling to meet current demands.

SG deserves credit for their work and the sheer quality of their hardware; even though they may be playing catchup right now, they have my wishes for their success. Every now and then I fondle the Indigo (and many Suns, Sun being in a bad position too) with sheer admiration for being built to last in a field where so little is.

BTW, you can thank Linux for doing serious damage to all these guys - who'd have ever thought that a commodity PC could do nearly as much as a professional-grade Unix workstation? I am a huge Linux fan, but it must be remembered that making a transition from a proprietary OS to Linux is a *huge* deal for companies who've probably made most of their incomes on software and support.

Re:Good for SGI and Sun.

[xenoterracide]

Last time I checked the Linux Kernel natively support's 32 processor's and Sun Solaris supports 128. I'm not sure about Datacenter because I think anyone using microsoft for any kind of server or datacenter is an idiot.

Re:Good for SGI and Sun.

[Gothmolly]

Windows 2003 Datacenter supports 64 processors (Which is more than enough for a regular commercial application. In case you want more, instead of scaling up, you should be scaling OUT.)

Translation: 640K ought to be enough for anybody.

Re:Good for SGI and Sun.

[Anonymous Coward]

It's more like "640K is currently good enough for anybody"

Re:Good for SGI and Sun.

[garyboodhoo]

ouch! that comment re: the mega expensive SGI CD drive brings back bad memories. Although it wasn't my money I consoled myself by rationalizing that the logo was the most expensive part of the package. I mean, it (the logo) was 3D and everything! Used to love my Indigo2 - great machine (like an uber-Amiga), great OS. Don't miss it one bit though. Really never understood how SGI managed to lose money when they overcharged for everything.

Re:Good for SGI and Sun.

[Anonymous Coward]

It's more like "640K is currently good enough for anybody"

Which was the original qoute yes, once mistranslated from a German article. But then we wouldn't have all the fun over the quote, so the incorrect meme is kept.

Re:Good for SGI and Sun.

[Anonymous Coward]

Well just in case you haven't noticed, CyricZ is a master in the art of bullshittery. Like the pointy haired boss at your work he has no clue what he is talking about but says what is needed to get himself attention. Also note that he posts about 20 times a day meaning that 1. he has no job, 2. he has no life. So in conclusion: Ignore CyricZ he is a spamming troll that is critical of others yet brings nothing to the table.

Re:Good for SGI and Sun.

[jacksonj04]

Up to 64 cores in Windows Datacentre Edition.

Re:Good for SGI and Sun.

[Thaddeus]

When did you last check, 1995?

Linux Kernel v2.6.14.4 Configuration
...
[*] Symmetric multi-processing support
(8) Maximum number of CPUs (2-255) (NEW)

I'm not sure it's a good path.

[jd]

It would seem to make more sense to have 4 virtual cores, where all have access to all computing elements. That way, you don't put such tight limits on the hyperthreading (where that would be beneficial). It would also be much easier to add more virtual cores, as it would not require any additional physical computing elements, so massive parallelization in cases where very different behind-the-scenes work would be needed would benefit further.

(Here, I'm using the term "virtual core" to mean a complete physical set of internal registers for each "core", but with no directly attached computing elements. Hence it is not a physical CPU, as it doesn't process, it's not central and it's distributed rather than being a unit. If you had a memory pool for such a purpose, where what was left was available as L1 cache, you could dynamically change the number of cores to suit the work you were doing. That would seem to be the "ultimate" design.)

Re:I'm not sure it's a good path.

[gimplar]

Sounds like you're describing the ultimate reconfigurable computer. Something that can , on the fly, reconfigure itself and change its architecture to the needs of the user. You're right, this would be great. Massively parallel and reconfigurable chips DO exist yet they are far from mature (http://www.mathstar.com./ If you're suggesting that the architecture should just magically morph from parallel to more parallel..I think you're asking a bit too much.

Re:I'm not sure it's a good path.

[dfghjk]

Ah, you mean a task switcher in an operating system. Great idea.

Re:I'm not sure it's a good path.

[Anonymous Coward]

They make these great things called FPGA's...

Re:I'm not sure it's a good path.

[Anonymous Coward]

You just described sun's Ultrasparc T1.

When will Microsoft change its license?

[TSHTF]

Currently Microsoft charges per CPU, not core http://www.microsoft.com/licensing/highlights/mult icore.mspx. As we begin to see 4-core and 8-core CPUs, how long will it be until Microsoft begins charging per core?

Re:When will Microsoft change its license?

[cyberjessy]

This is silly. Microsoft made a conscious decision to license software per CPU (or per Slot) rather than core, and they had announced that they are doing so because multi-core looked like the natural way in which CPUs will improve, given that the Mhz war has ended. In fact they were the first major company to do so.

Also, this does not really eat into MS bottom line compared to Oracle or IBM. Most of MS revenue comes from the the desktop, while they are just competing in servers. Sql Server suddenly becomes more attractive, given Oracles complicated multi-core policy. (Remember that Oracle earlier announced that every core is a CPU, its just recently that they realized it will be a disaster and modified their original plans.)

Earlier CPU speeds doubled every 18 months. Multi-core will simply take another approach to achieving the same. I am not sure how this will hurt software companies any more than increasing cycles/sec.

Re:When will Microsoft change its license?

[Deluge]

As we begin to see 4-core and 8-core CPUs, how long will it be until Microsoft begins charging per core?

Ah, but why would they? They'll just bloat the OS so it eats up enough CPU to require a multi-multicore chip setup for serious server performance.

Re:When will Microsoft change its license?

[quanticle]

/*As we begin to see 4-core and 8-core CPUs, how long will it be until Microsoft begins charging per core?*/

As I recall, earlier there was some confusion about Microsoft's license agreements regarding the core vs. CPU issue in terms of licensing.  Microsoft came out and explicitly stated that it was going to charge per CPU.

Re:When will Microsoft change its license?

[fbg111]

Earlier CPU speeds doubled every 18 months.

To nitpick, it's not speed that doubles every 18 months, but, roughly, the number of transistors per die. Clarification here.

Re:When will Microsoft change its license?

[Anonymous Coward]

Yes, that is Moore's law says. But if you read his post again you'll note that he made no mention to Moore's law.

It's like razorblades

[Anonymous Coward]

This time next year, we'll have 16 core CPU's from AMD to go along with our 16 blade razors from Gillette.

Re:It's like razorblades

[Hextreme]

It's amazing how bad "educated" people can be with grammer and punctuation. The apostrophe and comma are probably the two most misused punctuation elements - but seriously, if you're going to post on slashdot and have anyone take you seriously you need to learn to use them properly. CPU is an acronym for Central Processing Unit. When you pluralize that phrase, you add an S to the end of unit. What you wrote is "Central Processing Unit's" ... and those CPUs sure aren't owning anything in what you said.

Re:It's like razorblades

[digitaleden]

It's spelt grammar.

Re:It's like razorblades

[Anonymous Coward]

Was that "grammar" that you were referring to? I've never heard of this "grammer" thing before.

Re:It's like razorblades

[Karma Farmer]

You should mind your P's and Q's. The GP poster was wrong, but your explanation is wrong too.

The plural of CPU is CPUs. The plural of C.P.U. is C.P.U.'s.

Re:It's like razorblades

[Anonymous Coward]

If you are going to whine about commas, learn to watch for run on sentences in the post. Now, I am going to go shoot myself for making a grammar post.

Re:It's like razorblades

[Duhavid]

Quick question:

Does this make you the Grammer Nazi?

Re:It's like razorblades

[Anonymous Coward]

Even the grammar police haven't decided on a rule.

Re:It's like razorblades

[cerebis]

It doesn't matter in either case.

The pluralization of an acronym (a lazy man's construct) is purely based on common convention, and the use of an apostrophe is optional. After all, you've just dropped the word in favor of its leading letter, so why concern yourself with the trailing letter which could arguably be implied by context and included in the drop?

If you're frought with concern over using the correct plural form, just use the actual word instead or rewrite the sentence to not require a plural subject.

Re:It's like razorblades

[Limburgher]

Sorry. C.P.U.'s still refers to something belonging to the C.P.U. The periods don't change the grammar.

Re:It's like razorblades

[3770]

It's amazing how "educated" people fail to understand that many of the people on ./ aren't native English speakers.

The guy you were complaining about might be from China and highly intelligent. How good is your Chinese?

You better start practicing now though so that your Chinese is perfect by the time it becomes the dominant language on the Internet.

Re:It's like razorblades

[Anonymous Coward]

Wait... you mean punctuation is a matter of style, like brace placement in a C program? And you're saying that in some borderline cases, people don't agree on the styles?

Oh, the horror! The horror! How will we have "grammer" flames if the correct answer is "it's ambiguous?"

Re:It's like razorblades

[Daengbo]

That, of course requires the obligatory link.

Re:It's like razorblades

[Anonymous Coward]

If you can't figure these things out by context, you're clearly a fucking retard.

Re:It's like razorblades

[Anonymous Coward]

I suggest you have a look at Greengrocers' Apostrophe

In othe news...

[dcapel]

...scientists report global warming is predicted to increase four-fold by 2008.

Re:In othe news...

[goldstein]

Actually, the motivation for multicore processors is to reduce power consumption, or at least to limit the growth in processor power consumption that has been seen in the last decade. Intel's plans to introuced P4s with clock rates of 4 GHz and higher foundered, in large part, on the greatly increased power consumption that was required. The idea of the dual and multi-core processors is that a relatively small sacrifice in the clock speed for an individual core can result in a large reduction in power consumption. This results from the reduction in operating voltage that is possible at the slightly reduced clock rate and the additional reductions are possible if the deaign of the core is optimized for the lower clock rate. With the reduced power consumption, the performance gain of using two or more cores more than offsets the loss in performance that would have resulted from the lower clock rate.

Re:In othe news...

[dcapel]

Ye Gods man, it was a joke. I know it was totally off-base with reality, but that wasn't the point...

Analyzing jokes is generally a bad idea.

Re:DragonFly BSD will really start to shine.

[eviltypeguy]

While I don't doubt that DragonFly BSD will start to shine, I'm not sure that it will "take the lead." I personally expect OpenSolaris to take the lead since SUN has far more years experience in dealing with multiple processors.

So why do they go straight to 4 cores?

[Anonymous Coward]

Why do they go straight to 4 cores? Does it have to do with the architecture of hardware, or is it just another penis measuring contest?

Le-Yawn.....

[Groo Wanderer]

Wow, who would have guessed? Me.
http://www.theinquirer.net/?article=29550

Want to know what the problem is? Near the bottom here:
http://www.theinquirer.net/?article=25349
(Yes, I know I spelled it wrong, it was a verbal tip....)

            -Charlie

Re:Le-Yawn.....

[Anonymous Coward]

From your article

You know that it will be about a good chunk faster clock for clock, per core, than Dothan, especially on multimedia apps. Numbers like 30% are being passed around for encoding. Why? ... The other big one is an early exit for some instructions, and if memory serves, idiv is one. What this means is that an instruction can be retired after eight or so stages of the 13 in the pipeline. This can make a huge difference for things like encoding movies and the like.

Would this be why Apple went with Intel?

Re:Le-Yawn.....

[Groo Wanderer]

"Would this be why Apple went with Intel?"

It is part of it, but far from all. There is platforms, long term roadmaps, 'other' technologies, and an inbred dislike for Hector from his Motorola days. All play a part to one degree or another.

            -Charlie

Soon...

[LeddRokkenstud]

The amount of CPU's in a computer will out grow the amount of razorblades in a face razor!

new imac owner

[DarkClown]

posting this from the 2 ghz imac bought a couple weeks ago - i haven't had hardly problem at all since getting this (except for wish more apps were universal binaries).
the 'rosetta' apps actually do run faster than they did on my old 733 mhz g4, but when i found a new firefox build (beta) and replaced it and got tot a/b the difference it was really shocking - this system shines bright.

Pfft.....

[Anonymous Coward]

Only 4 cores and due out in 2007!? If multi-core is the future then sun must be ahead of their time:

http://www.sun.com/processors/UltraSPARC-T1/index. xml/

However,

[diksel]

"The Intel iMac Is Almost As Fast As The Quad Core Power Mac." Intersting results, from macspeedzone

Re:However,

[Jeff DeMaagd]

I think that is more of an issue of software not utilizing the available capacity, the software won't be optimized unless there is sufficient hardware available to justify optimizing the code. That very article seems to indicate that a Quad G5 could encode two videos at a time and still beat the Intel iMac.

Re:However,

[Yartrebo]

I thought about this before, and then came to the realization that cache and memory performance are the limiting factor in video encoding (specifically the snow codec [part of ffmpeg] behaves as such, but any wavelet codec or codec implementing overlapping block motion compensation should perform similarly). The asm code that I'm writting for the snow codec is almost twice as fast per clock on 32-bit AMD chips as it is on my 1.8GHz P4, with the tiny L1 cache being the primary culprit, but other memory issues and the long pipeline contributing too.

As Intel's multiple cores share cache and memory resources, they are unlikely to encode video much better than single core (at least snow video, which is very demanding, though not nearly as much as dirac).

This is pure hype

[kbolino]

With the limited number of mainstream applications that actually use multiple threads, the dual core CPUs are currently under-utilized. So, the natural response is therefore to double the number of cores, which will require additional licensing for Windows, and won't even be used by any (end-user) programs.

Re:This is pure hype

[east coast]

If you don't build the hardware the software will never be developed for it.

Re:This is pure hype

[kbolino]

You make a valid point.

Re:This is pure hype

[mrchaotica]

So run more than one program at a time, genius! Not all of us are still using DOS, you know!

; )

Re:This is pure hype

[4e617474]

So, the natural response is therefore to double the number of cores, which will require additional licensing for Windows, and won't even be used by any (end-user) programs.

So how is it that WinXP users buy a single-license copy, install it on a dual-core machine and find that the Task Manager includes the ability to limit a process to one core or the other? And how is it that people are actually using it to troubleshoot an end-user app?

Re:This is pure hype

[kbolino]

Didn't think ahead before I posted it.

Re:This is pure hype

[kbolino]

Windows XP Professional supports up to two CPUs. Not four.

Re:This is pure hype

[mrchaotica]

Actually, it wasn't a bad comment -- I was just trying to be funny.

Re:This is pure hype

[Yehooti]

Seems like with games it's the other way around.

Re:This is pure hype

[Asztal_]

Windows XP Pro allows two physical processors, which could well each have 4 cores.

Re:This is pure hype

[4e617474]

They still charge per installation on the desktop and CPU on the server not by the core. And there weren't dual core desktop machines when XP came out, there's no reason they won't be behind the curve by more than their usual margin when it comes to quad-core aware Vista versions. But hey, it's not like I've never said anything without thinking, I'd be posting at 1 ;)

In other news

[Burning1]

In other news:

Not to be outdone by AMD, Gellete releases a 5 core Razor.

Re:In other news

[19thNervousBreakdown]

News from the future!

Re:In other news

[anishm]

Actually, AMD and interl are the ones trying to catch up to gillette :) http://money.cnn.com/2005/09/14/news/fortune500/gi llette/

Re:In other news

[glitch23]

Rule #1: When making a joke, make sure you can spell.

Re:In other news

[Anonymous Coward]

And, as usual, The Onion is way ahead of everybody: http://www.theonion.com/content/node/33930

Re:In other news

[Phroggy]

OMG it's revolutionary brand new technology! We've... added one more blade, just like we did last time. And the time before that, with the Mach3.

Re:In other news

[Burning1]

Rule #1: Try not to take yourself too seriously.

Re:In other news

[Burning1]

That's exactly what I had that in mind when I commented. = )

Re:In other news

[glitch23]

Rule #1: I take myself seriously, just not the public (US) education system.

Well, it _would_ be...

[mrchaotica]

...except that neither SGI nor Sun has the foresight or initiative to grab the opportunity, and never will. Sure, it's sad, but it's true.

Die-mensional Computing

[Doc Ruby]

As I recall, the original Pentium (I) chip had 6 direct ports for direct CPU/CPU interconnect, forming cubic arrays (with PCI buses on the "surface"). Not only allowing scaling the CPU cycles, but also allowing tasks to "route around" unaccessible (crashed or broken) CPUs. Does the Linux kernel take advantage of such a CPU "network topology"?

Now that cores are going multiple, from double to quadruple, the next step could be octuple. That could offer a "cube" of processors in a single CPU, with similar redundancy/failover routing. Possibly addressable among the cubic CPU network. Is Linux able to exploit *that* kind of multiparallelism?

Re:Die-mensional Computing

[LLuthor]

Actually Linux is far more sophisticated than that.

Linux can efficiently handle NUMA and hotplugging CPUs at the same time. Memory locality and extremely scalable TLB management also help make it one of the fastest and most scalable systems out there. Linux can handle 128 CPUs out of the box as it were (vanilla kernel), and with patches from SGI up to 512 CPUs. I think only Solaris/SPARC can beat it with such high-end configurations.

Re:Die-mensional Computing

[Katate]

Cube's have 6 sizes, yo.

Re:Die-mensional Computing

[xenoterracide]

If it can't I doubt M$ or anyone else can right now, and It would be the first to be able to.

Re:Die-mensional Computing

[Rob Simpson]

And 8 corners, which I think is how he meant it.

Re:Die-mensional Computing

[Anonymous Coward]

The SGI patches should have given you a hint... Namely, IRIX can handle up to 2048 processors in a single system image. High-end multiprocessors are just about the only area where they are still on the top of the heap, but unfortunately the market for such beasts is limited to a few government agencies, and cannot sustain the company.

Re:Die-mensional Computing

[porl]

actually, if you look here you will see that you are in fact STUPID EVIL and the cube has only FOUR sides! :)

Re:Die-mensional Computing

[pimpimpim]

Reminds me of the Connection Machine 2, which had a cubic architecture from both the computing and the outer design: http://mission.base.com/tamiko/cm/cm-image.html

Vista

[BradWarden]

Is this the min required for vista?

Re:Vista

[Jozer99]

Assuming this is a serious question, the answer is one core. Vista makes use of multiple cores, but there are not necessary. For instance, Vista will run on the Athlon XP 3200+ (single core) or Pentium 4 3.2 GHz (single core, hyperthreading) just fine.

Re:Vista

[Belseth]

Is this the min required for vista?

That isn't the scary part it's the terrabyte of ram it requires.

Cell versus 8086.

[Anonymous Coward]

If you analyse the Cell architecture, you can see a an integrated circuit with a lot of simple processors inside.
Are these intel 8086 processors designs reaching their end because they are very complex to allow the building of multicore 8086 circuits with a low cost like a Cell processor?

Re:Cell versus 8086.

[mrchaotica]

No, because a modern x86 is just a RISC chip with a CISC-to-RISC translator and a shortage of registers.

Actually, it is a good path...

[mrchaotica]

... because it's the only choice they have. With your "super-hyperthreading" idea, the maximum IPC of the processor is still exactly that of the single processor. Hyperthreading may make a single core more efficient (in some cases), but it won't actually make it any faster.

In contrast, a dual-core chip has exactly double the maximum IPC of a single chip, regardless of hyperthreading.

Correction:

[mrchaotica]

I meant instructions per second, not per clock cycle. Whoops.

Octacore

[drix]

Why wait? Sun already makes processors with 8 cores. For realz.

Re:Octacore

[Courageous]

But they are weak cores... based on the old USPARC 1's. Just not the same as a quad core x86, which will hand one of those T1000's its lunch.

C//

Re:Octacore

[calidoscope]

But they are weak cores... based on the old USPARC 1's.

IIRC, the cores are based on the US-IIi's as used in the Ultra 5 and Ultra 10. They are also coupled with a crossbar switch to 4 banks of DDR2 RAM.

Re:Octacore

[twiddlingbits]

VERY VERY wrong,the T1000/T2000 family is the Ultrasparc T1 (note the 'T') chip which has Chip Multi-threading. If you want to think of it a different way think of it as an US VI chip. However, you don't want to use the T1000/T2000 for floating point. All 8 cores currently share a pair of FPUs so for compute heavy tasks use the US IV+ or Opteron based servers.

Re:Octacore

[Courageous]

I'm not wrong at all. Looking at the specjbb2005 figures for a T2000, which costs $16K typical, I can tell right now that if you compare it to a 4 CPU / dual core / 8 core opteron box, it will be completely destroyed. That box will cost half as much, too.

There is an issue of power consumption, which may just save sun's bacon. But that's the only reason.

C//

Re:Octacore

[twiddlingbits]

You got to remember the specJbb2005 tests are NOT designed to be extremely multithreaded and most of the tests run on 1 JVM. I can run 32 simultaneous threads on a T2000 and only 8 on an 4 CPU 8way Opteron/Xeon. The Opteron clock speed is about 2X. IMNSHO, the specJBB isn't giving a fair reading as in the specWEB the T2000 wins big. On the specWeb the T2000 (8 cores, 1 chip) scored 14001. I looked at the results for a IBM p550 with 4 Dual Core CPUs (POWER 5 chip) and it made a 7881. A Dell 2850 with 4 dual Core Xeons made a 4850. Another way to look at the T1 chip is as a 9.6GHz chip (8 * 1.2GHz). If you are really curious go to Sun.com and get your free 60 day eval on a T2000. I bet you'll be amazed.

Sun Microsystems is shipping 8 core CPUs

[Anonymous Coward]

Sun has been shipping 8 core T1 processors for a few months. T1000 and T2000 rack servers based on "CoolThreads" - whatever that is. See here: http://store.sun.com/CMTemplate/CEServlet?process= SunStore&cmdViewProduct_CP&catid=141649. IBM has QuadCore Power5+ chips shipping any day now.

Both Solaris and AIX scale over 100 CPUs already. Good luck AMD and Intel on getting Microsoft to create a standard OS (not their funny datacenter version) that is the same on 1 CPU or 124 CPU systems.

Re:Sun Microsystems is shipping 8 core CPUs

[Anonymous Coward]

Full forward binary compatibility across server OSes is very important. Sun has binary compatibility since Solaris 2.3, maybe earlier - meaning you can take a program (binary) that was compiled on Solaris 2.x and copy it to a Solaris 10 SPARC server (Ultra4, Ultra4+, or T1 based CPU) and run it unchanged successfully. See here: http://forum.sun.com/thread.jspa?threadID=27689&ts tart=75

There is a caveat about device drivers, but the vast majority of code **will** run fine. I'm not certain if IBM on RS6000, PowerX chips also is binary compatible, but I'm positive that HP-UX isn't after years of having to recompile old HPUX code to get it to run on the next major OS release.

Re:Sun Microsystems is shipping 8 core CPUs

[twiddlingbits]

Cool Threads is named for the T1 multihtreading chips that run a LOT cooler than the Hyperthreading (which is a poor imitation of Multi-Threading) Xeons. The T1 chip consumes about 65-70 watts versus over 100 for a Xeon. Internal chip temps are also not as high meaning you don't need as big of a heat sink and massive airflow (or liquid cooling) for the chips.

Re:Sun Microsystems is shipping 8 core CPUs

[Anonymous Coward]

Hey, math geek....

2^x == 124? Nothing. Oh well.

2^6 == 64; 2^7 == 128.

SGI is about to go belly up

[maynard]

and Sun isn't far behind. IRIX 6.x was great technology five - seven years ago, but it's old hat now. And Solaris x86 is no Solaris/SPARC. MS should have no trouble scaling Windows up to four - eight cores in the next iteration. The only issue is what they will want to charge. Linux is already there.

SGI and Sun aren't the winners here, I think....

Re:SGI is about to go belly up

[Jeff DeMaagd]

MS should have no trouble scaling Windows up to four - eight cores in the next iteration.

They shouldn't because they are already well beyond that with their Datacenter Edition:

http://support.microsoft.com/kb/888732/en-us

Re:SGI is about to go belly up

[maynard]

32 procs. Well, that's certainly enough for a desktop over the next few years.

Re:SGI is about to go belly up

[Anonymous Coward]

Actually, no. Sun is quite far behind SGI in the race to bankruptcy considering the billions of dollars they have in the bank and, now, very strong products.

Solaris x86 is no Solaris/SPARC

No, Solaris x86 and Solaris SPARC do share the same codebase and Solaris on x86 is hardly immature.

MS should have no trouble scaling Windows up to four - eight cores in the next iteration.

Right, except they haven't quite figured out how to scale up to one core yet.

SGI and Sun aren't the winners here, I think....

More cores per CPU from AMD and Intel really means the industry is accepting improved throughput, not clock speed, as where the next generation of CPUs need to go. This stands to benefit Sun since they are quite focused on this idea--in software and hardware. Their 8-core Niagara processors, which debuted in December 2005 are leading the pack in throughput, power usage, and rack space.

I don't expect x86 multi-core processors will be able to compete with the low power and space requirements of a SPARC multi-core. And by the time we have 4-core x86 processors in 2007, I would expect Sun to have upped the ante.

Re:SGI is about to go belly up

[jcr]

MS should have no trouble scaling Windows up to four - eight cores in the next iteration.

Yeah, but that's what, another eight years off?

Linux, Mac OS, Solaris, the BSDs, and even the remnants of BeOS will be doing a better job on 8-way machines before MS can patch Vista to even recognize them all.

-jcr

Re:SGI is about to go belly up

[Anonymous Coward]

You are an ignorant idiot. Windows NT's performance scaled decently on an 8-way system back in 1998, and the datacenter edition supports 32 processor configurations right now. More specialized versions are available for larger systems, I believe.

Re:SGI is about to go belly up

[Anonymous Coward]

Not very well.

Only like 70% or 80% gain per additional cpu compared to 99%+ of all the various unixes mentioned.

Re:SGI is about to go belly up

[jcr]

You are an ignorant idiot.

No, I'm an experienced customer who knows better than to believe any MS advocate's claims of performance, securability, etc, etc. Fool me once, shame on you. Fool me twice, shame on me.

Windows NT's performance scaled decently on an 8-way system back in 1998,

Would this be the microsoftie definition of "decently", as in "windows takes a decently long time between bluescreens these days"?

and the datacenter edition supports 32 processor configurations right now.

Heh. It's a long way from booting on a 32-way machine to "supporting" it. For windows to run on a 32-processor machine, it's rather like a dog walking on his hind legs. You don't even ask if he does it well, it's just impressive that he even attempts it.

-jcr

Per-core memory controller?

[chiph]

Someone call me when they offer a memory controller per core.

Chip H.

Re:Per-core memory controller?

[Bing Tsher E]

And seperate address/data busses, then? Sounds like a pretty high pin-count package.

Re:Per-core memory controller?

[quanticle]

/* And seperate address/data busses, then? Sounds like a pretty high pin-count package.*/

AMD pulls it off with their DirectConnect architecture...

Re:Per-core memory controller?

[Anonymous Coward]

It's all marketing. It's become the new mhz. By most accounts, AMD's engineering for their multicore chips is MUCH better than Intels. Rather than become a core race, Intel should redesign their current line of chips. Why won't this happen? It has been known for quite awhile that AMD chips outperform Intel chips for the most part. I think Intel pretty much has given up on actually having better chips (for now) and is now trying to make chips that SOUND better. Four cores sounds better than two. It may be a lazy sloppy design that won't scale, but at least it sounds good to the consumer. They did the same thing with their 64 bit chips.

Who knows, it may work. The reason Intel is so big isn't because of their amazing engineering, it's because of their amazing marketing. They've gotten away with inferior chips for the past 2-3 years and still remained strong, why should the immediate future be any different?

moore's law

[jotux]

So are the number of cores in processors going to double every 18 months?

Re:moore's law

[thisislee]

That actually does seem reasonable. IIRC, Moore's law doesn't say that computers will double in speed every 18 months; It says that the number of components on a chip produced at minimum cost will double every 18 months. I'd imagine double the amount of cores takes about double the amount of components(but likely a little bit more), so Moore's law probably does work for that. If computers get twice as fast is more subjective and depends on how programs are written.

Casual reading

[sdfad1]

A bit off-topic? I found some recent readings about the future of processing technology, and developments in software to exploit this, to be quite interesting.

Man, I gotta go find out why multithreading is so hard. Gotta try and get back into it someday and have a play around, get to truly understand it. I've only played with "Intro to Threads" type programs in Java, so there should be plenty for me to learn, when I get there...

In the mean time, there's a lot of things to look at for someone lazy like me. If I'm not gonna do them, I can read about them first. My favourite starting points have got to be (many Lisp related): Bill Clementson's blog. Erlisp (because I don't know Erlang itself, so Lispy Erlang would be a good way to understand the concepts I feel). Short explanation here. There's Erlang itself, but don't forget all the others: MapReduce, Termite (quite simple & beautiful code) and cl-muproc (all on the second link above).

Once (is it if or when?) single processors riding on Moore's Law stop working, is there no escape from concurrent programming? Looks like Erlang (currently being "borged" into Lisp, yay) is where the action is gonna be if massively multithreaded architecture turns out to be the winner for parallel/concurrency/distributed computing.

Re:Casual reading

[Jeremi]

Man, I gotta go find out why multithreading is so hard.

I'll tell you why it's hard: "Normal"/Single-threaded programming is like training Helen Keller to dance: not easy, but once she's learned the layout of the room and memorized the steps, she can do the routine pretty well. Multi-threaded programming is like training a dozen Helen Kellers to do a synchronized group routine: each of them still has to know the steps and avoid tripping over the furniture, but they each also have to know how to do it without ever colliding with any of the 11 other dancers. Add to that the fact that every dancer moves at a unpredictable, varying speed with respect to the other dancers, and you've got quite the challenge if you want to have a successful show night after night.

Still, it can be done, and it can even be fun once you get the hang of it. Prepare for a good bit of hair pulling and "why does my program crash randomly, about once a week?" until then, though...

Re:Casual reading

[fjf33]

Have you ever tried to take a dump and pee at the same time? Similar problem. Not really, what really is the main problem is that most programs run in sequential steps and there is limited paralelism. For example, if you are trying to make a cake you cannot start cooking it until you mixed the ingredients, however you can mix the batter and the glaze at the same time so there is limited paralelism. That kinda gives you an idea of the time/improvements from two cores. It is like having two cooks. They also get in each other way so there is some efficiency loss there. Overall for most tasks is really not much gain when you look at it per task. However if you are cooking dinner then you may be running two or three dishes at the same time, very little in common bu they don't have to wait for each other. There you gain a LOT from multiple chefs. Of course one may be sitting idle waiting for the other to release the oven. That is the resource contention problems a core faces when trying to get something say like a byte from memory. If you have two ovens (or two memory controllers each accessing a different bank of memory) then you can cook at the same time with little contention for resources.

Re:Casual reading

[Tiger]

Thank you, this is one of the nicest (and most amusing) analogies I've seen for describing the difficulties of concurrency. :)

Good for Linux

[r00t]

Right now, SGI is using Linux on the 1024-way systems. Thanks to careful planning and IBM's help with the RCU patent, Linux scales way up. (and down too) Linux had the advantage of former Sun and SGI developers telling horror stories about what NOT to do.

Windows represents CPUs using the bits in a machine word. On 64-bit hardware, you're limited to 64 CPUs. This is exposed in the ABI. Not that Windows would scale well for such a system, of course.

Apple seems to be behind as well.

Re:Good for Linux

[Anonymous Coward]

Windows can be partitioned into multiple instances on machines with more than 32 CPUs. Thus windows can take advantage of hardware such as that, albeit not in a single system image.

Re:Good for Linux

[glens]

> windows can take advantage of hardware such as that

It rather sounds to me like what you're saying is that it's such hardware that can take advantage of (or, overcome the limitations of, really) Windows; not the other way around.

Re:Good for Linux

[drsmithy]

Windows represents CPUs using the bits in a machine word. On 64-bit hardware, you're limited to 64 CPUs. This is exposed in the ABI. Not that Windows would scale well for such a system, of course.

Why not ?

Specific, technical details with references, please - not just rhetoric.

Bandwidth

[Aardpig]

I wonder whether the quad-core Intel chips will be as bandwidth-starved as the dual-core chips? Currently, the comparison between a dual-core Pentium and a dual-core Opteron is farcical, especially for memory-limited apps.

Re:Bandwidth

[MojoStan]

I wonder whether the quad-core Intel chips will be as bandwidth-starved as the dual-core chips? Currently, the comparison between a dual-core Pentium and a dual-core Opteron is farcical, especially for memory-limited apps.

The quad-core chip that Intel demoed in TFA, code name Cloverton, is derived from the Merom/Conroe/Woodcrest architecture (which is derived from Yonah/Core Duo). It is not derived from the current bandwidth-starved Xeon core (Netburst/Pentium 4).

I agree that the comparison between dual-core Pentium D/Xeon and dual-core Athlon 64/Opteron is a farce. However, Yonah (even with its 667MHz bus and non-integrated memory controller) did quite well against Athlon 64.

Of course, that doesn't mean bandwidth won't be a problem for Cloverton when they move to 4 cores. I believe Cloverton will still have a FSB and seperate memory controller while Opteron will have next-gen HyperTransport and integrated memory controller(s).

Re:Bandwidth

[the eric conspiracy]

Yohah v. Athlon comparisons are not really applicable here. These are 4 way server parts with different operating requirements and characteristics than desktop applications.

New from Gillette?

[stavromueller]

For some reason I'm thinking Gillete Quadro...
So how many blades/cores do you really need????

OpenSolaris and DragonFly won't take the lead

[r00t]

IBM licensed the RCU patent for GPL software. Well, if Sun really does use the GPLv3, maybe they could use RCU. DragonFly will be GPL when the BSD devil gets to ski and ice skate.

Solaris helped Linux to scale, mainly by showing what NOT to do. Back when Linux was starting to get serious about SMP, the design was strongly influenced by horror stories from former Sun developers. Solaris and IRIX suffer from excessive locking. The locking is so complicated that it causes developers to add new locks instead of using ones that ought to be used, which only makes things worse.

Re:OpenSolaris and DragonFly won't take the lead

[laffer1]

Well Sun has changed the threading design several times in Solaris. You need to be more specific. Sun used a model where one cpu was the controller (scheduler) and the other cpus ran jobs in sun os 4.x and in early Solaris versions it used a m:n model where there would be m user threads to n kernel threads similar to how Microsoft's .NET framework creates threads. Newer versions of solaris (9 and 10) are more like linux and freebsd's latest threading library and make 1:1 relationships between userland and kernel threads. I'm not an expert at this and have been taking an operating systems class this term where this has been discussed.

As for dragonfly, I do think that Matt was right about some problems with freebsd 5 and 6 but each release is getting faster. 6.1 beta is noticably faster. Dragonfly isn't revolutionary though. I think some of the ideas from Mach inspired some of their design decisions and we all know Apple has the most succesful Mach kernel in the commercial world.

I don't know if we'll see freebsd or dragonfly look super impressive on multicore cpus, but I do know that openbsd and netbsd may not scale well depending on what they are working on. I can tell you that freebsd 6 does fine on my dual xeon machine. Solaris 10 on the same system seemed slightly slow but i think that was driver support more than anything. Linux IS SLOW on the system regardless of the scheduler. For that OS class, I had to install the 2.6.15 kernel and custom compile it for my system prior to our work on adding system calls. Its not as fast (gentoo) as freebsd 6 was on the system, but faster than freebsd 5.x. (especially disk io) I don't know why linux seems slow as it is using both cpus quite well. Of course this is percieved speed.. i haven't done any formal benchmarks.

Maybe someone should do a serious benchmark on FreeBSD 6, Netbsd 3, Dragonfly (whatever the latest is), OpenBSD 3.8, Linux 2.6.15 (gentoo distro?), and for kicks OpenDarwin all running on the same dual core hardware. Hell if i get time this summer, i might do it.

Re:OpenSolaris and DragonFly won't take the lead

[Anonymous Coward]

Dragonfly has a very long way to go. At present it is all talk and no walk. It doesn't even perform up to FreeBSD, and we all know that FreeBSD scales poorly. Realistically, Dragonfly is still a proof of concept project, essentially hoping to copy its basic design features from Amiga0S.

Re:OpenSolaris and DragonFly won't take the lead

[r00t]

You're right about the way Solaris threads work, but I was thinking more about the way the kernel accesses its own internal data structures.

Has FreeBSD suddenly changed course? Their threading history as I know it is: pure user-level thread hack as the default for version 4 (most systems start this way), a LinuxThreads port as an option for version 4, and now adding a very buggy exotic system for version 5. The exotic system is somewhat like what Solaris used to do, generally being called an N:M design.

Matt seems to have taken to heart the advice that Larry McVoy was giving the Linux developers about a possible way to avoid the "locking cliff" that SGI and Sun went past. The Linux developers responded to Larry's warning with awareness, education, and paranoia. (following the normal path, but VERY carefully)

Drivers are always a problem for benchmarks. Probably you should benchmark on both AMD64 and EM64T, with two different SATA controllers, in both heavily-loaded and lightly-loaded situations. Be sure to include at least one test that involves disk IO and much more than 4 GB of RAM. Another problem is variable integrity behavior. Not even counting ext2 and the exotic filesystems, Linux (ext3) has three different journalling choices.

Re:OpenSolaris and DragonFly won't take the lead

[laffer1]

FreeBSD has 3 thread library choices as of 5.4 and 6.0. libc_r, an M:N implementation and a newer 1:1 library. Only the 1:1 library is in active development and patches are not usually applied to the other two. The 1:1 library was developed during early FreeBSD 5.x development for kernel operations and after benchmarking it, someone made the call to push it into userland.

Recently, on one of the freebsd lists (hackers or performance) there were some benchmarks with mysql on the different threading implementations. I think that was a good sample of the performance characteristics of each if you're interested.

A quad system bus of some kind would help more

[CFD339]

Processing power is rarely the problem. Graphics processing is already handed off, and unless you're trying to crack encryption, most software isn't bound by processor speed anyway.

Software performance is bound by I/O limitations. It FEELS like processor power because threads on hold for I/O block a core up like cheescake to a lactose intollerant grandparent.

Until I can index on disk at about 100 times the current speed, these processors won't help what I'm doing.

Re:A quad system bus of some kind would help more

[WuphonsReach]

I dunno what you do for a living (or a hobby), but there are those of us for whom quad-core won't get here soon enough. Disk I/O is always going to be an issue, but there are a lot of applications that still need more CPU power.

I keep a dual-CPU Opteron system pretty busy most days and I'll be upgrading to a dual-CPU dual-core system once prices drop a bit more.

Re:A quad system bus of some kind would help more

[Anonymous Coward]

It FEELS like processor power because threads on hold for I/O block a core up like cheescake to a lactose intollerant grandparent.

Ummm. Lactose intolerance does the exact opposite of blocking them up. It means it shoots outta their ass faster than they can get to a bathroom.

Re:A quad system bus of some kind would help more

[timeOday]

most software isn't bound by processor speed anyway... software performance is bound by I/O limitations.

That's a ridiculous overgeneralization. Just because you're a database guy doesn't mean we all are.

Re:A quad system bus of some kind would help more

[Anonymous Coward]

The new pricing for the dual cpu capable Opteron chips has just hit, and the only question now is going to be availibility. I assume the demand for these things is going to be outrageous for the next few months. (The 265 is the same price as 165 ?!) Now to find a place to put my old 246's...

(More specifically, for ~$1K, you can put together a dual cpu, dual core (4 total) 1.8GHz machine, now that AMD dropped the prices of the CPU's by half)

Re:A quad system bus of some kind would help more

[lubricated]

>> most software isn't bound by processor speed anyway.

I'm sure the software development you do is very boring and will be outsourced.

Re:A quad system bus of some kind would help more

[2008]

Games. They don't touch the hard drive unless you're loading a level.
They're often limited by the graphics card, but more CPU can do good things too. I live in hope that all the computing power from multi-core CPUs will lead to an AI opponent in an RTS game that plays like a human, and ideally individual units (thousands of them!) that have a similar intelligence to a human too. Not having to babysit everything would actually allow strategy, instead of tactics.

Re:A quad system bus of some kind would help more

[Anonymous Coward]

Oh yeah. You're totally right. It's the HARD DRIVE'S FAULT that it takes several hours to compress video. Perhaps some day someone will invent a hard disk capable of reading and writing faster than 250k per second, allowing amazing gigabyte per hour video encoding speeds! It's so silly how all that CPU power is going to waste, blocking on storage hardware that hasn't improved since the mid-80s.

Or, returning to the REAL WORLD, we will see that in fact there are plenty of common CPU-bound tasks for which I/O throughput is not a contributing factor towards completion rate.

Go ahead, index your disk as fast as you want, but don't act like your databases are somehow important enough to dismiss the larger category of, y'know, COMPUTING.

Dual core systems are great..

[cowmix]

Dual core is the best thing to come down the pike in years.. I have always opted for a dual processor system over a 'faster' single processor box because for the type of things I do, run a zillion things at one time, having multiple processors made everything run very smooth. The only issue I had with recent dual processor machines I have owned is the intense heat and power consumption...

Dual core processors solved that problem.. I'm golden.

Race is already over

[Midnight Warrior]

This multi-core race is already over. Sony wins by default with the PS3 coming in with 7 cores (#8 is a ghost to cover over manufacturing flaws and defect counts). And everyone is whining about how to code for 7 cores. Having four cores won't change this single-threaded world. When the libraries of the world are suddenly multi-threaded, the PS3 will be light years ahead. Plus, IBM is going to be putting their Cell processors on blades. IBM and AMD are two years too late to the game.

Re:Race is already over

[SirTalon42]

Sun has a processor with 8 cores. Also the Cell doesn't have 7 (or 8) 'cores', and IBM did most of the development of the Cell, with the help of Sony and Toshiba.

The Cell is also not a general perpose processor, its rather focused on a few things (like image processing).

Re:Race is already over

[Anonymous Coward]

I don't think the race is over yet... especially because, SUN already launched last year a 8-core chip (Niagara), I have one at my office and it rocks ! It is a very simple chip and our software (agressively multi-threaded) flies on such hardware. Niagara chip has 8 cores, and each core runs 4 threads in parallel, thus it can run 32 threads in parallel, this is just amazing ! I don't think Sony is so ahead, in this race. Build a 8-core chip is pretty easy. Having softwares that take advantage of it, might be more difficult. I doubt that all PS3 softwares will fully take advantage of it anyway.

Re:Race is already over

[Anonymous Coward]

The T1 isn't a general-purpose processor then, either. It's a memory controller with some meager integer support and comical float capabilities.

Re:Race is already over

[Anonymous Coward]

That's wrong. Each core can schedule four thread contexts, but can only execute one thread context per core per cycle. When a stall occurs in a thread context a sibling context is scheduled in its place when possible. The actual computational facilities of the T1 are meager (with numeric code being its antithesis), and it's only interesting in the context of moving a lot of data around in transaction processing. Neither the Cell nor the T1 are in the same "race" as other multicore processors. They're specialty hardware.

1 core = 1 frying pan -- 4 cores = 4 frying pans

[elbonian]

So now we can fry four eggs at the same time? In 64 bits too?

Be more concerned about the bus

[Courageous]

This whole multi-core trend concerns me. Sun Niagra is now out, in the form of the Sunfire T1000 and T2000 computers. These are fine computers. But they really only excel for very specific workloads. Meanwhile, facts are facts. The chips are starved for data.

It's almost comical how the Slash community seems to be so back and forth over which chip is "best". Cart meet horse. Get behind, thee!

So. I am a bit of an AMD fanboi. I admit it. But it's not really about the chip. It's the IO fabric. Hypertransport (which does happen to be on chip) is why AMD is winning this race right now. It's affordable, and scales linearly with the number of chips. Around the corner on AMD's front is HORUS (http://www.theinquirer.net/?article=18251), the memory fabric to rule them all. Intel should be really afraid here.

I personally can't get all excited about these multi core chips. Now IO solutions. Those interest me.

Computers are entirely IO bound these days. Hello?

Do any Slashdot readers happen to be home in there!?

*knockety knock*

C//

OS isn't everything

[quanticle]

While operating systems that scale well certainly are important, I fear that not enough priority is being given to making properly multithreaded applications. Having a multithreaded OS can only take you so far without properly written applications to take advantage of those OS features.

Re:OS isn't everything

[ozmanjusri]

Having a multithreaded OS can only take you so far without properly written applications to take advantage of those OS features.

Virtualisation.

Re:OS isn't everything

[quanticle]

Virtualization isn't going to make the individual application run faster. 

Re:OS isn't everything

[Kristoph]

I am not sure I understand ...

The parent's point was that any given application cannot take advantage of the multiple CPU's. It is a given that if you are running mulitple applications, each of these would take advantage of each CPU.

What is the point of Virtualisation?

In fact, if you need Virtualisation to make use of mutliple CPU's/cores are you not loosing a tangible chunk of performance (not to mention a significant chunk of memory) running multiple OS instances?

]{

Re:OS isn't everything

[sgent]

Don't point out the obvious :) There are some rare cases that virtualization makes sense -- running multiple network servers where any one server doesn't have much demand, and you want them isolated from each other for security concerns. Virtualization is also useful in development environments where you can have an XP, Linux, SunOS, and 2000 running concurrently. That being said, I dont see virtualization as being the great advance in computing technology that everyone thinks it is. Its a cool toy, with a few important niche applications, but as you observed it often makes more sense to have 2 machines than 1 machine with 2 VM's.

Re:OS isn't everything

[jaseuk]

We have 100 servers, each of them has very low load. The reason we have seperate servers is to reduce management overhead (ie. one app one server), it's easier to rebuild and manage if there are not a load of different services all running on a single box, it's also harder to upgrade, restore and test if servers run too many services. In many cases outside contractors are involved with the maintenance and upgrades of individual applications, on a shared server they are likely to blame the other services leaving us in an awkward situation.

Of course having 100 servers to manage and upgrade brings it's own problems and this is where virutalisation has a niche.

VMWare claim for each CORE they can handle approx 8 virtual servers, in a production environment for reasonably loaded servers lets take this down to 4. With a dual-proc quad core machine we could comfortably host 32 virtual machines assuming that we can afford the RAM to host all this and that we have fast SAN storage. We could potentially reduce our server footprint from 100 down to 5 or 6 servers (including a few for resiliance) with this sort of technology available.

Also don't underestimate the benefits of a consistant hardware platform, snapshots, KVM, Power Cycling and imaging. That is something that VMWare can give everyone even if they host a single virtual machine per host.

Jason.

Adapted from the Onion

[amliebsch]

Fuck Everything, We're Doing Four Cores

By Craig Barrett
CEO and President,
Intel
February 10, 2006

Would someone tell me how this happened? We were the fucking vanguard of computing in this country. The Pentium 3 was the CPU to own. Then the other guy came out with a 64-bit x86 CPU. Were we scared? Hell, no. Because we hit back with a little thing called the Itanium. That's 64 bit and a new instruction set. For performance. But you know what happened next? Shut up, I'm telling you what happened--the bastards went to two cores. Now we're standing around with our cocks in our hands, selling 64 bits and a new instruction set. Floating point performance or no, suddenly we're the chumps. Well, fuck it. We're going to four cores.

Sure, we could go to two cores next, like the competition. That seems like the logical thing to do. After all, one worked out pretty well, and two is the next number after one. So let's play it safe. Let's make a faster bus and call it the Pentium4SuperExtreme. Why innovate when we can follow? Oh, I know why: Because we're a business, that's why!

You think it's crazy? It is crazy. But I don't give a shit. From now on, we're the ones who have the edge in the multi-core game. Are they what's inside? Fuck, no. Intel is what's inside.

What part of this don't you understand? If one core is good, and two cores is better, obviously four cores would make us the best fucking CPU that ever existed. Comprende? We didn't claw our way to the top of the processor game by clinging to the 64-bit industry standard. We got here by taking chances. Well, four cores is the biggest chance of all.

Here's the report from Engineering. Someone put it in the bathroom: I want to wipe my ass with it. They don't tell me what to invent--I tell them. And I'm telling them to stick two more cores in there. I don't care how. Make the wafers so thin they're invisible. Put some on the bottom of the die. I don't care if they have to cram the fourth in perpendicular to the other three, just do it!

You're taking the "point" part of "floating point" too literally, grandma. Cut the strings and soar. Let's hit it. Let's roll. This is our chance to make CPU history. Let's dream big. All you have to do is say that four cores can happen, and it will happen. If you aren't on board, then fuck you. And if you're on the board, then fuck you and your father. Hey, if I'm the only one who'll take risks, I'm sure as hell happy to hog all the glory when the four-core CPU becomes the computing tool for the U.S. of "this is how we compute now" A.

People said we couldn't go to 64-bit. It'll cost a fortune to manufacture, they said. Well, we did it. Now some egghead in a lab is screaming "Four's crazy?" Well, perhaps he'd be more comfortable in the labs at AMD, working on fucking Turions. HyperTransport, my white ass!

Maybe I'm wrong. Maybe we should just ride in Motorola's wake and make embedded IC's. Ha! Not on your fucking life! The day I shadow a penny-ante outfit like Motorola is the day I leave the CPU game for good, and that won't happen until the day I die!

The market? Listen, we make the market. All we have to do is put her out there with a little jingle. It's as easy as, "Hey, gaming with anything less than four blades is like playing at VGA resolution." Or "You'll be so l33t, I couldn't snipe you with an aimbot." Try "Your b0x is going to be so friggin' fast, someone's gonna walk up and put a goddamn spoiler on it."

I know what you're thinking now: What'll people say? Mew mew mew. Oh, no, what will people say?! Grow the fuck up. When you're on top, people talk. That's the price you pay for being on top. Which Intel is, always has been, and forever shall be, Amen, four cores, sweet Jesus in heaven.

Stop. I just had a stroke of genius. Are you ready? Open your mouth, baby birds, cause Mama's about to drop you one sweet, fat nightcrawler. Here she comes: Put another Level 2 cache on that fucker, too. That's right. Four cores, two caches, and make th

Re:Adapted from the Onion

[Anonymous Coward]

Very nice. But you missed one: "Hey, gaming with anything less than four blades is like playing at VGA resolution."

Re:Adapted from the Onion

[kireK]

We are Fucked
By Craig Barrett
CEO and President,
Intel
February 11, 2006

Some ass wipe in sales pointed out that Sun started shipping 8 core machines last year. To make it worse another fucking engineer pointd out that the Sun CPUs pull half the power that our dual core boxes do.

I quit.

Re:Adapted from the Onion

[amliebsch]

Very nice. But you missed one: "Hey, gaming with anything less than four blades is like playing at VGA resolution."

D'oh!

I mean...uh...he's talking about blade servers, obviously.

For those not familiar with the original: http://www.theonion.com/content/node/33930

Discussion by year 2342

[elbonian]

Dude one: I've bought a laptop with the new Intel FGHI 3xyz! Dude two: What does it have? Dude one: 3Gc (Gigacores) Dude two: Man! That is awesome! I'm gonna buy one of these. What power does it need? Dude one: A portable nuclear plant model NUKE32 or better.

Multi CPUs? Why not Cluster?

[Paraplex]

Why can't we take advantage of multi CPU software using networked PCs?

Re:Multi CPUs? Why not Cluster?

[Anonymous Coward]

There is different way to scale with softwares. For instance, you can scale per process or one single process can use multiple threads. Threads are "lighter" than process. You can not efficiently use threads across CPU, but inside a CPU it is very cheap. If a CPU has several cores, it can physically run threads in parallel. In a single core CPU threads do not physically run in parallel, the parallelism is simulated. Those multicore CPU open lots of new opportunities for standalone softwares. Sometimes it is impossible to split a software across multiple machines, because sharing information between each instances is just too expensive. You have it to run in one single machine.

Re:Multi CPUs? Why not Cluster?

[Ravatar]

Some people don't want 4-8 machines taking up space in their computer room? (As if they don't already)

I'd prefer 1 machine that can do the same.

Re:Multi CPUs? Why not Cluster?

[Jeremi]

Why can't we take advantage of multi CPU software using networked PCs?

Well, you can, for some things. The problem is that the speed of even gigabit Ethernet is so slow (compared to the speed of the interconnect between two CPUs on the same chip or motherboard) that in order for a program that runs well over a cluster, it often has to be designed completely differently from a program that will run well on a multi-CPU machine. In particular, things like shared memory are extremely difficult to simulate over a network, but trivial when you've got a single machine where the CPUs, well, share the memory.

For a lot of problem types, the high latency and low bandwidth of the network interconnect lead to the situation where you've spent hours and hours getting your app to run across all the CPUs in your cluster, only to find out that it's actually running slower than the simple version of the app that does everything on a single CPU. That's really embarrassing...

Says something about the seemingly endless advance

[cerebis]

The rate of the manufacturer switch to multicore CPUs has exceeded my expectations and really does say something about the seemingly endless advance of the CPU since its invention.

There have been many "the end is near" predictions over the decades, and none have come true. However, when the manufacturers start turning out dual and now quad core commodity parts, you really have to assume that a reasonably solid wall has been hit for once.

Though I have faith there are still depths in silicon to be plumbed, they obviously aren't close enough to fruition to be included on a manufacturers near-term roadmap.

As cool as it may be, parallelism has its limitations and is not the catch-all solution. There simply are some algorithms which do not lend themselves to it, though you could argue that alternative parallel friendly choices may always exist. What effect this parallelism pressure might have on solving problems, or choosing problems to solve, is something worth considering.

Even so, lets hope developers have some prescience about solving (or at least addressing) the problems inherent with concurrency and parallelism. I am already growing tired of listening to game developers working on new titles sidestep the issue and I've experienced some problems with older games which simply can't cope with two cores. Get writting those APIs. ;)

Great excuse...

[rfernand79]

Great excuse to keep promoting fatware instead of cleverly designed software.

Re:Great excuse...

[bprime]

Right, just like well-designed roads prevent good automobile design.

Besides, if the user doesn't notice it's bloatware, who cares?

Re:Great excuse...

[DeathFromSomewhere]

My God he's right! Everyone, let's all go back to programming in assembly!

SUN had 8 core CPUS in 2005

[kireK]

Sun is currently shipping EIGHT core CPUS, and each core handles 4 threads... so you are talking 32 threads in one RU of space.

http://www.sun.com/servers/coolthreads/overview/in dex.jsp

Not Surprising

[LionKimbro]

Intel said they were going to be releasing x100-core processors by 2015.

Just watch the Intel 2005 Keynote speech, (video) hear about x100 cores and x100 GBits/sec chip-chip data transfer.

It's not like this is a big secret or anything.

Re:In other news...

[jibjibjib]

...which is why you use it.

Re:Says something about the seemingly endless adva

[4D6963]

"There have been many "the end is near" predictions over the decades, and none have come true"

Well as you surely know, that's pretty much the magic of the Moore law. No matter what happens, CPU's will improve in performance (although the law really talk about transistors), period.

However, I don't know about you, but that multi-core thing gives me the feeling that they are getting into this because they're having a though time getting over 4 Ghz. As if, because they hardly can improve the CPU's the way they always used, they needed another solution to go on further. I mean look, new CPU's still come out with barely 2 GHz. Back about 10 years ago, you were getting from 66 Mhz to 200 Mhz in no time.

All Your Multi-Cores Are Belong To Us

[Byethoster]

All Your Multi-Cores Are Belong To Us lol :)

Re:All Your Multi-Cores Are Belong To Us

[Anonymous Coward]

The same content is available at http://www.processors.byethost11.com/spam_is_good, http://www.processors.byethost11.com/i_like_watchi ng_paint_dry and http://www.processors.byethost11.com/byethoster_is _a_pea_brain
The content in all cases are: irrelevant banner ads

Feel mean today? Then mod parent post to hell.

Obligatory...

[jheath314]

64 processors should be enough for anyone

Re:Obligatory...

[archen]

Microsoft press release for Exchange 12 release in a few months:

"Oops, we were wrong..."

Re:Obligatory...

[xenoterracide]

at one time vista's was supposed to be out in 2005 I believe. So an M$ Press release means nothing.

Re:Obligatory...

[Anonymous Coward]

Don't be so sure about that. My employer is rolling out an application that requires 300+ CPUs to work simultaneously. Predicting the weather ain't easy...

Re:Obligatory...

[gregeth]

I'm personally just waiting for someone to boot their OS off of their graphics card. :)

Ars Technica has a better article...

[Malor]

As usual, if you want good CPU info, Ars Technica is the place to look. They have a blurb on Intel's 4-way core plans here.

Basically, they point out that Intel's dual core processors are already starved on the FSB, and loading two more cores isn't going to do very much. He seems to expect that, until Intel gets their FSB in order(which won't happen until 2008), AMD is going to stomp all over them. He says that Intel's cores are excellent, but without CSI (their new FSB), it may not matter much.

My own projection is that the extra contention may end up imposing a net speed _penalty_ for many workloads. That is, however, pure speculation from an amateur, based mostly on the dismal performance of the first dual-CPU G4 Macs.

Re:Ars Technica has a better article...

[mattkime]

>>based mostly on the dismal performance of the first dual-CPU G4 Macs

What are you talking about? The first dual G4 Macs ran much faster than single processor machines - as long as you were running multiprocessor aware software. Unfortunately, when they were released not much was. OS X now makes great use of both processors.

Re:Ars Technica has a better article...

[jiushao]

I'd say that at this point it is much too early to claim that Intel is behind. As it currently stand their 1066 MHz FSB only manages to fill dual-channel 533MHz DDR2 granted, while there is fairly affordable 667MHz DDR2 out there (and not-very-affordable 800MHz), the Conroe which is not that far off now is expected to bump the FSB to 1333MHz, filling dual-channel 667MHz DDR2. This may then sound like a bad thing seeing how there is already faster memory out there, and AMD's hypertransport links scaling to insane speeds. As it happens though the AMD M2 processors will only have support for dual-channel 667MHz DDR2 ("only", it is pretty damn ridicolously much bandwidth), making them just match the capabilities of the Conroe.

Additionally DDR2 incurrs higher latencies than regular DDR, which is an advantage for Intel (although small) since the low-latency memory controllers on Athlon64's will have slightly less impact that way.

So: While Intels FSB situation might not have much headroom, while AMD has ample opportunity for bandwidth but no integrated controller that can use it. So the fact is that the whole of AMD's advantage is that they will be alone having 667MHz DDR2 between socket M2's release and the 1333MHz FSB Conroe, whereas Intel will have been alone with 533MHz DDR2 up until the release of socket M2 athlons.

Re:Ars Technica has a better article...

[Anonymous Coward]

"As usual, if you want good CPU info, Ars Technica is the place to look" - by Malor (3658) on Saturday February 11, @10:02PM

You mean arstechnica, a site online that has 'authors' like Jeremy Reimer who write about the field of computers?

Sure, a real 'authoritative site', not.

Example given is Jeremy Reimer:

Jeremy Reimer, an arstechnica article author, doesn't even possess a degree in computer science.

Jeremy Reimer, an arstechnica article author, doesn't even possess practical hands-on years of professional experience in computer sciences as a programmer or network engineer??

Jeremy Reimer, an arstechnica article author, who doesn't even possess a certification in the field of computers (not even an A+ or MCSE)???

Arstechnica = the home of the wikipedia plagiarizers is more like it.

Arstechnica is the home of quacks like Jeremy Reimer who enjoy posing as authority figures in this field.

Reimer doesn't apparently realize that people often look into the background of those who write articles to be wary of the fact that the person writing it may be a fraud like Reimer is.

Don't trust arstechnica. They have kids with no experience and idiots like Reimer writing their articles. Nothing original or fundamental in the lot. Simply regurgitations of already known knowledge via plagiarism and spouting back what they read elsewhere.

Re:Ars Technica has a better article...

[Anonymous Coward]

Sounds like a mac fanboy got cockslapped.

Hey, programmers

[Anonymous Coward]

I don't believe most modern software products work any differently on multi-core systems. We need multithreaded programs that actually utilize dual-core before we should be interested in quad-core. Oh, and imagine a Beowulf cluster of these.

How many have seen / used the Sun T1 processor

[GuyverDH]

Up to 8 cores, 3MB L2 Cache (total shared), 4 execution threads per core, so effectively 32 execution threads per CPU.

A nicely loaded Sun T2000 system, with 8 cores, 32Gig RAM, Dual 2GB FCA and 8 Gigabit Ethernet interfaces comes in with a street price of approximately 30K. Add in Solaris 10 with it's container technology, the fact that it only uses 325Watts of power, and is light on BTUs - we're talking serious datacenter contender for web services, application servers, database servers, etc...

I'm currently looking at consolidating approximately 20 aging systems using over 10KW of power and close to 20K BTUs/hr thermal output. I am planning on replacing these 20 systems with 4 T2000 servers totaling 1500KW and approximately 5K BTUs/hr thermal output. Not only will I be saving on maintenance for the hardware, but also on software licensing as common applications like Oracle and BEA are licensing their products at 1/4 cpu per core on these processors. I will also be saving on power and cooling requirements for the datacenter. Not to mention datacenter floor space - I will be able to empty 2 full racks with this consolidation project. I'm hoping to expand it and end up with 1 rack of T2000's replacing close to the entire datacenter's UNIX population.

Re:How many have seen / used the Sun T1 processor

[puddingpimp]

Would someone please mod this down.

Re:How many have seen / used the Sun T1 processor

[GuyverDH]

LOL - It appears that someone forgot their medication this morning.

Those numbers I quoted are straight from the environmental spec sheets. The same ones that every company develops to be used by datacenter planners to determine how much cooling capacity, as well as power requirements for everything running in the datacenter.

As for Sun systems being slow, that was mostly a by-product of their design. In the past, Sun systems were designed for heavy workloads - not speedy, but massive workloads, that wouldn't bring the box to it's knees.

Now, Sun has taken the time to re-write their OS, many components from the ground up (like their TCP stack) as well as their primary filesystem - the ZFS, to be released with the 6/06 Solaris 10 distribution.

These boxes are not slow by any means. Add to that, the processing capacity they have designed into the OS and processors, and you'll find that when you combine power, form factor (2RUs) and capacity, you won't find much that matches them.

I was primarily trying to point out, however - that Sun has 8 core here - NOW. While Intel and AMD are *talking* about 4 core next year.

I personally use AMD at home, but at work, I need something that gives just a bit more.

Re:How many have seen / used the Sun T1 processor

[Anonymous Coward]

parent poster = teh n00b

Re:How many have seen / used the Sun T1 processor

[DeathFromSomewhere]

servers totaling 1500KW

You may want to consider building a small power plant.

Re:How many have seen / used the Sun T1 processor

[GuyverDH]

LOL - I knew that didn't look right.

I was trying for either 1500W or 1.5KW - guess who fumbled the keys? =D

The one thing I don't do, is video transcoding...

[CFD339]

...other than that (which admittedly will dim the lights), I've never kept even a hyperthreading P4 pegged solid for more than a short time. I've written multithreaded audio processing software than can pick DTMF or Motorola QC-II tones out of a live audio stream with good reliability, and I've written anti-spam software that in test processes more than a million messages per hour. In neither case did I overwhelm the processors.

Neither of these tasks is important for a desktop user. Large software compiles, scientific analysis -- these are specialized tasks and sure multiprocessor machines will help. The market for these, however, is specialized and limited.

IMO - there will be external specialized processor blocks for these kinds of tasks. For example, an external FFT processor board would make a great deal of sense. Its a well known algorythm, used in a huge variety of analysis -- and is easy to predefine with parameters. Think of a PCI-Express board with its own set of RISC processors dedicated to performing FFT transforms in the same way modern graphics work is offloaded. You define in the drive the frame size, domain size, etc etc.. and get back the results as fast as you can pass them in.

The point is, what holds back solutions like this the most -- is the front side bus.

Not to quibble, but I think that depends....

[CFD339]

...on the person, no? I'm no doctor, was just looking for a funny line to relate to "blocked up" -- oh well. As Steve Martin said...."Everyone THINKS they have a sense of humor." :-)

Re:Also OT...

[Karma Farmer]

Are you saying Angry Flower doesn't have the final say on punctuation?

One of the RCU patents is in the public domain

[Anonymous Coward]

You can check the Linux RCU docs for which one. There's also other lock-free algorithms in the public domain. So far not a lot of interest. I think it's all still too new. It will take a lot more processors before the pain level becomes enough for everyone to consider other approaches.

Enough about the chip.

[Statecraftsman]

What we need are more brains.

The only way to begin dealing having 4+ cores in the cpu is to develop higher-level languages and compilers that can distribute and parallelize load optimally across the silicon.

Maybe these multi-core's can speed up interperted languages like Python, Perl and Ruby or be designed to specialize for services like httpd, mysqld, etc.

Another good point I saw raised on this issue is on having the IO to feed all these cores. The biggest problems being HD and FSB.

Re:Enough about the chip.

[tomstdenis]

This is pure bollocks. Any half-way competent programmer can start a thread or two and pass it information via globals and locks. I wrote in all of 5 minutes a dual-thread RSA accelerator for my cryptolibrary that effectively halved the latency of RSA private key operations.

In many "industrial" applications they're just different processes. E.g. an apache server hardly needs to spawn two threads to handle a single request. But having two threads for two requests make sense.

In desktop cases most of what you do or see can be threads or processes. Sound, graphics, gui, io, networking, etc... So an OS can make good use of multi-core/processor setups too.

Only an mutt thinks it's significantly hard to at least make some use of multi-execution designs. Now I agree making Optimal usage is hard but even then a programming language would hardly help you do something like stripped NUMA...

Tom

Re:Enough about the chip.

[Anonymous Coward]

We already have Dual Core brains

Oh, we're behind the times.

Re:Enough about the chip.

[Anonymous Coward]

Even without each program being multithreaded, we can still benefit by having multiple singlethreaded processes running simultaneously.

For example, watching an MPEG video in one process while burning a music CD with another process. Both processes can be singlethreaded and the OS will have the primary thread of each automatically run on a different cpu or core.

People who like to run multiple programs simultaneously will see an immediate benefit to having multple cores or cpus.

I suspect that games, virtual machines (vmware), and scientific applications will be the driving force in leveraging multiple cores. Business apps don't have to bother because GUI frameworks tend to use threading behind-the-scenes so that your average clueless programmer can write responsive event-driven programs.

Re:Enough about the chip.

[ClamIAm]

I'm sort of wondering how the CPU is going to evolve. Will the cores stay equal in power, being carbon copies of each other that rely on the OS to do everything? Or will we eventually have a hierarchy of processors in our system, with one (or a few) that is specialized in coordinating what's happening, while the others, specialized for execution speed, grind away on code (a la Cell)?

It would be more the reconfigurable...

[jd]

...that you were thinking. There are many different places you can parallelize, in a CPU, and many different forms of parallelization. It would not be that there was "more" parallelization (overall), merely that you were diverting resources so that you have more hardware where you need it, and less where you don't. The sum total would remain the same.

The 8-core Alpha

[Animats]

Just before the end of the Alpha line, DEC prototyped an 8-core Alpha. It's not clear that it was useful, but they did succeed in cramming eight CPUs on one die over five years ago.

Re:The 8-core Alpha

[thunrida]

Alpha was never multicore. EV8 had 8 instructions wide ooo execution, which was probably amazing, but it has nothing to do with multicore.

Re:The 8-core Alpha

[Anonymous Coward]

It had 8 execution threads on a single core, not multiple cores. It was the poster child of symmetric multithreading (SMT), the technology that was crippled by Intel to produce "hyperthreading". At least now we have Sun holding up the development of SMT with its new 4-thread 8-core processors.

142 replies and nobody posted it yet...

[symbolset]

Ok, time to sacrifice some karma.

"Imagine a beowulf cluster of those..."

ah, I'm too slow. It seems they already exist

[CFD339]

For example, Blackfin makes one.

Presumably, you get some kind of software driver to talk to with your own software so you can feed it the time domain in real time and get back blocks of frequency domain data without burdening your processor.

If you're doing serious scientific analysis on live analog data streams (SETI anyone?) this would really help you out a lot. I'd bet one of these could really jump your SETI@HOME scores up pretty quickly.

http://www.eg3.com/WebID/dsp/fft/

Err, no.

[jd]

Because you'd be fetching from N instruction queues in parallel, you'd be capable of running N times as many instructions at the same time. (You've only got one bus to memory, so your "fetch" doesn't change, no matter how many cores or physical CPUs you have. Indeed, half the problem with SMP is precisely that - you can't have multiple CPUs fetch simultaneously.)

Just to make this clear, I am not talking about one physical CPU core with multiple registers it cycles between. That would be ghastly slow. I'm talking about the ability to fetch and execute multiple instructions purely in parallel, no cycling, but where you are not left with idle resources in any CPU core when another CPU core could really benefit from them.

(This does mean "tagging" all internal operations with which core they go with, so that results go to the right place. That's a slight complication. You don't need that when you've tight coupling, because there's only one place the results CAN go to.)

The idea is that, behind the user-programmable part, you've pools of compute elements that are kept physically isolated in a "pure" multi-core CPU. This is wasteful, if one of those CPUs needs compute elements that another CPU is leaving untouched. When doing SIMD, it's ok. "Single instruction, multi data" implies you'll be using the same compute elements on all CPUs anyway, so there's no benefit to having common pools. However, that leaves out all multi-tasking and all MIMD code. With those, the odds are very high that different tasks will have very different CPU requirements, so soft-coding what elements go with what core would be highly beneficial.

Each core should be different

[backslashdot]

At least for consumer processors, imho they should probably be moving towards chips where each core has it's own specialization, for example .. one core does the general CPU stuff .. the other .. I dunno specializes in graphics or physics vector math or is a GPU. Another core could orient towards all sorts of video/media stuff.

On my home computers, my cpu utilization is minimal. The biggest bottleneck is I/O.

I'm waiting for a reasonable cost computer that can drive multiple wireless lcd touchscreen displays/voice interfaces throughout a house. When not in use these wireless? usb lcds can show pictures or serene screensavers. Fairly large displays can be used to show HDTV or movies.

Re:Each core should be different

[vux984]

Fairly large displays can be used to show HDTV or movies.\

HDTV screensavers any way.

Because in this imaginary future HD-movies will be so wrapped up in legal and technological protections that the signal will not be passed by any electronics over 4 feet away from the physical disk, making remote viewing screens impossible.

In related fake-future-news, the projection market collapsed when it became impractical to view hdtv movies. Mounting the player on the ceiling next to the projector was just inconvenient and then their was still the legal grey area surrounding the fact that the actual screen was 10 feet away.

=p

Re:Each core should be different

[4D6963]

I smell optimism in the air!

;-)

Re:Each core should be different

[hyc]

Sounds like an Atari or an Amiga. Or a mainframe.

But yeah, you can get a lot of mileage from having dedicated coprocessors (including dedicated I/O processors). Something that everyone except Apple and commodity PC makers seems to already understand. Of course, there are still limits to that, and it's always nice to have a faster central processor.

Re:Casual reading / multithreading

[dfries]

Man, I gotta go find out why multithreading is so hard.

Why is multithreading hard? What's a MMU good for? Multithreaded programs are hard because there are multiple tasks running (which can be at the same time on multiple cpus), which share the same code and address space. If task 1 messes with the memory that task 2 is working on, then task 2 can cause a seg fault and kill the entire program. To a lesser degree that is the same problem with systems that don't have (or don't use) a MMU. The memory management unit and a modern operation system makes it so that one program can't screw with another program's memory (there are ways around that which debuggers use, but for the most part that's the case). Then again if you write perfect code without bugs you don't have to worry about the above problems.

Even if you could write bug free programs there still is the problem of how do you as a programmer divide up what needs to be done among the availabe threads? Can you logically divide up everything that needs to be done? Something like for an audio playback program, one thread to read data from disk and decompress it, another takes the decompressed data and feeds the sound card, while another updates your GUI or text output. If you were doing a raytracing program you could divide up the image into blocks and give each thread a block to produce.

You don't have to do multithreading to take advantage of multiple cpu or multiple cpu cores. For the audio example you could have one process decode the audio data and pipe it to another process to send to the sound card. Multiple processes will take advantage of multiple cpus just like multiple threads of one program. Same with the raytracing case. Heck you could even communicate between multiple computers rendering one frame if your inner process communication was done with TCP or UDP. Try moving one thread of a program to another computer. Won't work.

No, no switching.

[jd]

This would do away with all task switching and time slicing (up to the limit of the virtual cores). This would be executing instructions in a true parallel fashion, not by rotating between them, but it would (a) shift the CPU resource allocation from the OS to the CPU, and (b) eliminate all context switching. Instead of tightly binding resources to a given CPU and loosely coupling the context with the code, you'd loosely couple the resources to the CPU and tightly bind the context to a given block of code.

As context switching and time slicing are huge contributors to latency and cycle overhead, eliminating those should produce smoother, faster code. They are also heavily re-used bits of code, which means eliminating them should be kinder to the stack and to the kernel threading mechanism.

In any optimization, you always want to go for the biggest contributor that you can change significantly. These are probably the most frequently used parts of the kernel and this would not just simplify them but would eliminate them entirely.

(If you take this to the extreme, you'd start with a nano-kernel - so you've only the really essential stuff that absolutely can't be in userspace in kernel space - and push as much of that into hardware as you can. You can't push everything - yet - but the less you have in pure software, the less switching you need and the faster it'll run.

96 Cores

[yerdaddie]

As long as we're waving "number of cores" about as if it were the number of inches a piece of spam is promising:

http://news.com.com/Designer+puts+96+cores+on+sing le+chip/2100-1006_3-5399128.html

In short, Clearspeed's CSX600 has 96 cores, but is designed to be an accelerator board.

so there is no feeling good about this?

[SteveXE]

I just built a new computer specs will be posted below. I plan on building another in 2 years so I suppose by the time I do there will be a reason to own a quad core cpu but still...why do they have to make people who just got a new box feel like total ass by saying stuff like this...? Why not wait till a couple months before release to start their propaganda wars thus allowing people like me to be happy with my purchase? Sure I should be used to it by now but damnit im not! AMD 4800+ Dual Core Dual (SLI) nVidia 7800 GTX's (512mb) 2 GIGS Corsair XMS PRO DDR 1 TB Raid Array Liquid Cooled and Overclocked Now it all feels useless and im just gonna throw it in the garbage!

Re:so there is no feeling good about this?

[porl]

i'll have it if you don't want it. ;)

Front side bus

[peterfa]

Is it not the front side bus that chokes the whole system?

Re:Front side bus

[NerveGas]

Only with an Intel. A64/Opterons have an embedded memory controller, so the hypertransport link out of there only needs to feed the AGP/PCI/PCI-E busses, and it's got gobs of bandwidth for that.

Re:Front side bus

[peterfa]

Hmm, I'll get an AMD system then.

Why do you say that?

[autopr0n]

Proper punctuation is only valued by a small minority of readers. Someone can be taken nearly as seriously with extra apostrophes strewn about like so much of yesterday's garbage..

Those are slow

[jd]

At least, compared to a good, high-speed processor. However, some of the ideas from FPGAs, multi-path hardware, "fat trees" and packet switching are perfectly usable within a CPU to redirect internal operations and to provide common resources. It is not a crime to re-use ideas. (Ok, SCO thinks is, but I'm ignoring them for now.) As for the idea of having unequal division of labour, that's a big part of IBM's "cell" technology but is really as old as "intelligent" peripherals and the use of dedicated processors. All I'm doing is taking existing methods and looking at what mixes make for the best gumbo.

There is absolutely bugger all point in eliminating time delays from a lack of resources if you end up adding even greater delays by soft-coding everything into an FPGA. What's wanted is to hard-code everything that is concrete (so you run at maximum speed) but soft-code everything that's already abstract (so you don't introduce delays through implementation that were never a part of the design).

Mine goes to 11

[ruiner13]

My chip's cores go to 11. Take that.

Re:Mine goes to 11

[Anonymous Coward]

Hey! So does the volume knob on my slightly used Spinaltap amp! No drummers were spontaineously incinerated with this equipment though....

if only...

[digitallysick]

if only software would catch up and support dualcore, and 64 bit.

Well, yeah.

[jd]

Actually, it describes an idea that's been around since at least the late 70s, which is why I'm surprised that the reaction has been so negative to my post. (It wouldn't be so bad, if they'd pointed to an example as to WHY this wouldn't work.) My only dispute with these existing approaches is that either the technology didn't exist at the time (the late 70s was a trifle early in the parallel era to do a good design, and they couldn't have created a CPU with enough transistors to do the job right, even if they had such a design) or they are not taking enough advantage of everything they could do, presumably because it's fashionable to do as much as possible in software.

What I'm looking for is a way to take the trends being followed to see where you would logically end up if chip designers keep following them. My guesses are probably way off what the "end result" would be, but do seem to be in the general direction of where people are going, so would seem a good starting point for speculation.

so does this negate moore's law?

[Rooked_One]

I mean, becuase instead of having the transistors double, it looks like cores per chip are doubling.... so i guess moore is half right?

Re:so does this negate moore's law?

[dabraun]

Right, because putting twice as many cores on the chip isn't going to double the number of transistors. The cores are only, what, 10 transistors each? The rest of the transistors are just there to satisfy moore's law, not to do anything useful.

Re:so does this negate moore's law?

[lasindi]

becuase instead of having the transistors double, it looks like cores per chip are doubling.... so i guess moore is half right?

Well, if the cores doubled every year and the cores had more or less the same number of transistors, that would mean that the number of transistors double ...

Re:Also OT...

[mmd]

However, I will grant that the same argument for insisting that "begging the question" means raising the question can be used to dismiss the difference between acronyms and initialisms. Does that make it right? Well I guess that depends on if you are one of those proscriptivists (who insist that supposably is a word as long as enough people use it) or a descriptivist (who insist that words mean something, and ivory tower dictionaries contain those absolute meanings).

Why, no. That depends on whether you [sic: real pedants use "one"] are one of those proscriptivists...

Jeez. Half-assed pedantry is so unsatisfying.

Skip four, go to five!

[Eric Smith]

Fuck Everything, We're Doing Five

Re:Skip four, go to five!

[n6kuy]

I'm waiting till they go all the way up to 11!

Re:Skip four, go to five!

[tomstdenis]

This sounds funny until you realize that companies like AMD [and Intel I imagine] have fallback plans for 3-core processors in case 1 core is dead on production.

So it's entirely possible when the next series of 6 or 8 core processors are made you can have in theory anywhere from 1 to 8 cores active. Chances are they won't allow chips with >1 defective core to be sold for quality reasons. But it's entirely possible that you could buy a 5-core processor by 2010 :-)

I really hope these quad-cores involve 32 or 64-bit HT links to cope with the bandwidth problems. Only so far you can push HT links. Currently with multi processor setups it's a bunch better than Intel's offering [I know this because I work with a 4P dual-core Opteron setup at work...] but even two cores can push an HT link south. ... Wants the 4P box for home ... drool

Tom

Re:Skip four, go to five!

[steveoc]

640 cores ought to be enough for anyone.

I understand what you're saying, however....

[CFD339]

What I think you may be missing, is that with a stronger FSB or some similar path to external tools, the processing power you want can be added more easily and cheaply as purpose built asic processors -- RISC chips with specific purpose.

An example I gave earlier is FFT for scientific work (converting analog signal data over time into frequency blocks). This algorythm is the same for a huge amount of tasks, is very processor intensive, and is common but used widely for different things. IOW, perfect to offload from the processor.

A really good FSB path that didn't block processing other things, could solve your processing power issues much more effectively with a plug in hardware solution to offload.

Use of these kinds of purpose built processing circuits is limited by the FSB. Graphics cards work well, but remember, the ultimate output is to the monitor, not back to the system. The calculation is passed through the bus, but the result is not. Even with that compromise, bus speeds limit the solution so much that AGP and PCI-E have been invented as a way around the lack of a really fast general bus.

Just imagine if you could offload that heavy calc work to a purpose built card -- or potentially a whole additional computer, what the result would be for performance.

3 cores for malware detection

[kantmakm]

and 1 for actual programming. I don't like this math.

Re:3 cores for malware detection

[Anonymous Coward]

Here's a nickel, kid...

Re:Oh the childishness

[bussdriver]

Clearly the next big hype from intel will be around multi-core cpus. Even less to gain than from there waste for high clock speed fad. (now they are running smarter, at lower rates.)

Nearly everything is starved for bandwidth. The real gains naturally are low bandwidth well designed code for multiple threads. MOST consumers will not be running that, and with an advanced GUI like OS X, the work is being dumped on the GPU more each revision... Making the extra cpu less influential than it was before 10.3 and 10.4. If you have tasks that can properly exploit the 4 cpus, then it IS worth it.
I was slowed down by the RAID, RAM, BUS before the Quad G5 and its even more clear with 4 cpus.

Why wait? Cost.

[CyricZ]

Of course Sun already makes processors with 8 cores, in addition to numerous systems with far more processors than that. But those cost a lot. Far more than most individuals could possibly afford, and far more than many small businesses are willing to spend.

A dual 4-core processor system based on this technology from AMD may prove to be far more affordable. Thus it becomes a system that is within the financial grasp of individuals, and a real bargain for business users.

Re:Why wait? Cost.

[Anonymous Coward]

a T1000 costs under 3K. did you even bother to go to the sun site to check the price!! its cheaper than any x86 both and you get 8 physical and 32 virtual cores

Re:Why wait? Cost.

[CyricZ]

The lowest end Sun Fire T1000 server costs £ 2,200.00. It doesn't come with a hard drive. You can get an 80 GB SATA drive from them for £ 81.94, but that's a hassle, and often isn't enough for many uses these days. Also, it only offers 1 PCI-E slot. It will be great as a server once it's available, but it's not desktop material.

Even the higher-end T1000s only include an 80 GB hard drive. While external storage is always an option, it isn't something that most regular users wish to deal with.

One can get an entry-level Ultra 20 for £ 560.00, and yes, it comes with a hard drive. Even a higher-end Ultra 20 will run you £ 1,700.00. A low-end Ultra 20 costs £ 1,450.00. Keep in mind that those are actual workstations, where you can easily add and remove hardware.

The number of CPUs is one factor. But there are many other concerns, including storage capacity and the ability to add additional hardware. Sun could offer a package including the best of both worlds, and they could potentially rival Apple.

Re:Why wait? Cost.

[Anne Thwacks]

Can I get an 8-core Niagra processor as an upgrade option for my Ultra 20?

We are Unix users - we run the windowing system on a separate machine anyway, so we dont care if the T1000 has no graphics or HD. A HD is only for boot and swap space - real storage is elsewhere. (Doesn't every home have a bank of Raid servers these days? You will need them if you are going to have a network based media centre - especially one using MS software.) Since a T1000 will boot over the network, just stuff it full of RAM and forget about swapping!

This WAS written on Sun kit running FreeBSD).

Re:DragonFly BSD will really start dying.

[Anonymous Coward]

It is now official. Netcraft confirms: *BSD is dying

One more crippling bombshell hit the already beleaguered *BSD community when IDC confirmed that *BSD market share has dropped yet again, now down to less than a fraction of 1 percent of all servers. Coming on the heels of a recent Netcraft survey which plainly states that *BSD has lost more market share, this news serves to reinforce what we've known all along. *BSD is collapsing in complete disarray, as fittingly exemplified by failing dead last in the recent Sys Admin comprehensive networking test.

You don't need to be the Amazing Kreskin to predict *BSD's future. The hand writing is on the wall: *BSD faces a bleak future. In fact there won't be any future at all for *BSD because *BSD is dying. Things are looking very bad for *BSD. As many of us are already aware, *BSD continues to lose market share. Red ink flows like a river of blood.

FreeBSD is the most endangered of them all, having lost 93% of its core developers. The sudden and unpleasant departures of long time FreeBSD developers Jordan Hubbard and Mike Smith only serve to underscore the point more clearly. There can no longer be any doubt: FreeBSD is dying.

Let's keep to the facts and look at the numbers.

OpenBSD leader Theo states that there are 7000 users of OpenBSD. How many users of NetBSD are there? Let's see. The number of OpenBSD versus NetBSD posts on Usenet is roughly in ratio of 5 to 1. Therefore there are about 7000/5 = 1400 NetBSD users. BSD/OS posts on Usenet are about half of the volume of NetBSD posts. Therefore there are about 700 users of BSD/OS. A recent article put FreeBSD at about 80 percent of the *BSD market. Therefore there are (7000+1400+700)*4 = 36400 FreeBSD users. This is consistent with the number of FreeBSD Usenet posts.

Due to the troubles of Walnut Creek, abysmal sales and so on, FreeBSD went out of business and was taken over by BSDI who sell another troubled OS. Now BSDI is also dead, its corpse turned over to yet another charnel house.

All major surveys show that *BSD has steadily declined in market share. *BSD is very sick and its long term survival prospects are very dim. If *BSD is to survive at all it will be among OS dilettante dabblers. *BSD continues to decay. Nothing short of a miracle could save it at this point in time. For all practical purposes, *BSD is dead.

Fact: DragonflyBSD is dying

Did you ever use SGI systems?

[CyricZ]

Of course such capabilities have been available to paying users for decades. Nobody is suggesting otherwise. What is being suggested, however, is that these capabilities can be economically brought down to everyday users. Apple is a favourite for mid-range multimedia applications, a market that SGI once dominated. Indeed, if SGI were to put out a lower-end system like the Indy again, things might must pick up for them.

I take it from your general attitude that you never seriously used SGI systems. They were lightyears ahead of their time, and in many respects aren't even equalled today. Their proprietary systems, while expensive, were of an extremely high quality. Unlike today, where you buy a Dell and know it'll likely fuck up within a short time span, SGI systems were engineered to function, and to function for ages.

Lately we've seen near-death companies turn themselves around. Apple managed it, and Sun isn't far behind. It would not be unreasonable to see SGI make a comeback, assuming they're quick enough to take advantage of opportunities like this. They need to start focusing on lower margin products, and offering affordable systems featuring 8 or more cores could be a good start.

hmmm

[Anonymous Coward]

Can we have a better race? Might I start with a few sugestions- Race to more complex poorly supported hardware, well that race seems to be hotly conteded with nvida and apple. For flaimbate potential: the opensource race for who can make a Linux or BSD that can hoorkup your system the fastest. The current winner is Gentoo closely followed by Arch. Or howabout who can wine rant and complain the the most about "expensive labour" ? Well hard to say but who's winning that one. It's close call right now. As for hertz go, as others have noted- I perssonolly don't see anny "real" benifits . I admit though I have become fairly conservative though. My old comparitively old PC and Macs work just fine for anything I care about.

Re:Says something about the seemingly endless adva

[AuMatar]

Its about time we need software to pick up the slack. For the past 2 decades, we've been counting on improvements in hardware to improve our software performance, and eating up much of that improvement in bloat, inefficient algorithms, featuritis. The worst came recently with the rise of languages like Java where you're giving up 20+% performance just by language choice. While there have been some improvements coming out in algorithms, it hasn't nearly kept up with hardware, or with bloat.

If we want future increase in performance, it needs to come from software- better algorithms, better parallelism, and real effort by developers to achieve this. If we don't get this, we've basicly hit a wall in computer performance. The only big increase to come is drops in RAM latency, and I don't see that around the corner.

terrible article...

[YesIAmAScript]

It speaks of the dated-ness of Intel's FSB. The problem isn't Intel's FSB, in fact it has more useable bandwidth than AMD's right now (until DDR2 AMDs come out).

The problem Intel has is how the two cores communicate to each other (mostly for cache coherency). Intel's cores do this over the FSB, AMD's do not.

You may say I'm splitting hairs here, that Intel's FSB may have the same or better bandwidth, but it is hogged by inter-CPU communication (unlike AMD who uses HT). But I would beg to differ, if you understand the problem, you realize that once you understand the actual problem, you wouldn't put in a statement like "As I pointed out in a post on Intel's major 2006-2007 weak spot, these four-core processors will be crammed into a socket that's fed by an out-of-date front-side bus."

It isn't a problem with the socket. If Intel fixed their cores to communicate with each other directly instead of over the FSB, they would be in essentially the same position as AMD, and they could do it without changing the socket, since the connections would be internal.

All this FSB stuff is a little bit confusing, because AMD doesn't even actually have an FSB. Athlon has a memory interface and an I/O interface, but no classical FSB. The FSB is normally used to reach the memory controller, and AMD has that on chip.

Re:terrible article...

[alonso]

I think you assume that the next generation AMD cpu will use only one "bus" to comunicate with the ram, like current generation of opteron and X2. I think that the next socket with 1200 pin will permit to link directly core with memory. This will be make the difference.

Possible Ad Campaign?

[Anonymous Coward]

The new "slick intel quadro" quad processor!

Be careful to benchmark your app first...

[trims]

I work at Sun, in what used to be called JavaSoft. Just so there's no claim of hidden bias.

The current generation of T1/2000 (called "Niagara") has a major limit, which may or may not impact your app. There is only a SINGLE FPU for each 8-core die. And that FPU doesn't implement the full IEEE floating point specification in hardware.

Every time you need to do an FPU call, it's a major performance hit. So be careful of the app you run on a Niagara-based system. The last word I heard from the Engineering dept was that you wanted to make sure NO MORE than 10% of your instructions were FPU-bound, and that you really should keep the mix to under 5% for best performance. So if you plan on running a home-grown app on it, you might consider source-code changes to eliminate as much floating point math as possible.

All this said, the Niagara makes the conscious decision to be FPU-limited. The market it aims at runs very FPU-light applications - I've seen it run Apache, and boy, does it smoke any x86-based machine when doing so (particularly if your site has heavily database-driven content, in a typical 2-tier setup). Very Fast, Inexpensive, and really Power-Efficient.

Just make sure you ask for a demo machine to test your app with (or talk with a Sun Field Engineer first), so you can find out if Niagara is really for you.

All of this is publicly-available information (go look at the technical documentation Sun has released on the Niagara chip), so I'm not giving anything away that isn't known. Just pointing it out to people to make sure they're happy with the product they buy (we don't want people buying a Pickup and claiming we were supposed to sell them a Ferrari).

That said, the next-generation Niagara (codename "Rock") is already in the works. Talk to you Sun sales rep (or call the 800-number) to find out what's in the works for it, since I can't discuss it here.

-Erik

Re:Be careful to benchmark your app first...

[GuyverDH]

Aye, I should have mentioned the FPU limit myself. However, since most of the apps we run are FPU light apps, we don't notice the limit.

I can hardly wait to see the reported improvements in the Niagara 2. 1 FPU per core, plus the SSL acceleration that's already in each core. Add in the increase from 4 execution threads to I believe 16 execution threads per core. I believe it will be in 2, but it might be in 3 where they'll add in the scout chip to pre-load contents into the cache just before execution time so each thread has it's cache loaded with most if not all the stuff it needs.

I am not a Sun employee, I am just a Sun customer, who for the first time in years, is truly enthused about the release of an OS level and processor that seems to make a lot of sense.

OT: On pedantry.

[David Rolfe]

First, a note to mods: This comment is already marked off-topic in the subject line. Why you would need to double mod it I have know idea.

Why, no. That depends on whether you [sic: real pedants use "one"] are one of those proscriptivists...

Jeez. Half-assed pedantry is so unsatisfying.

Your comment seems to be even more niggling than my fun-fact (as you are solely making a suggestion of style rather than a correction). Assertion if x and/or y would never require whether. I'm not aware of any grammar in which whether was requisite in a conditional. "Depends on whether" is certainly a common contruct (proscriptivists cheer!) but is no more semantically valid than "depends on if".

Finally, if I were writing an essay, and not merely replying to an individual, I would have certainly used more formal constructs (viz. pronoun choice). Still, the choice between if and whether in a conditional is not a grammatical choice.

Is nitpicking style "half-assed pedantry"? If that's the case, let me give you some style advice: Real pedants use one. Words mentioned as words are not quoted, they are emphasized, typically with italics. Also, 'real pedants' is a classic example of the Real Scottsman fallacy.

Thank you for the reply.

Re:OT: On pedantry.

[randyest]

Why you would need to double mod it I have know idea.

"Know" idea indeed. This must be a troll. I have no idea how one could seriously try to be a pedant and fail so miserably. Most morons wouldn't even try; you must be a special kind of moron.

Finally, if I were writing an essay, and not merely replying to an individual, I would have certainly used more formal constructs (viz. pronoun choice).

Would you also use two unnecessary commas in your essay?

Better make a high end applications...

[Clazirus]

If the trend start to make more and more core processor available, then sofware developer better making a high end application that capable utilising the core processors performance.

It's not OS what matters, it's applications

[Anonymous Coward]

And I know which one exactly is going to really shine :)

AMD has demonstrated quad-core already.

[Anonymous Coward]

But have they demonstrated working quad-core CPUs the way Intel did in TFA?

AMD has long since demonstrated quad-cores behind closed doors (NDA). The news here is that Intel decided to demonstrate openly. The difference is in the amount of PR it generates, not the time to market.

Four cores good

[kars]

Four cores good, two cores baaaad!

Re: SUN had 8 core CPUS in 2005

[tomstdenis]

Don't mistake the # of cores with their efficiency. An AMD64 core can execute upto 9 instructions per cycle (though really that's just theoretical) and you're looking at 3 at most in practice (do to decoder bandwidth).

Where the AMD64 multi-core shines by comparison to the Niagra chip is the per core cache and more advanced out of order execution engine. Niagra is a very task specific chip whereas AMD and Intel chips are more general purpose.

This whole "4 threads per core" issue is just bollocks anyways. If each core only has one ALU pipe then who cares how many threads you have. Specially since they share caches where you will hit a huge bottleneck.

Tom

Multicore vs Single vs Hyperthread vs Multi CPU

[Whiteox]

There really is a big choice out there now, but many don't realise the significance between these types.
Very generally speaking, Hyperthread (or similar) allows switching between active apps. eg it is easy to switch from an app that is rendering a video file, to the desktop or another app. Normally a computer would wait or at best lag behind while it's looking for processing cycles.
With a 2 core, it is possible to do the same as hyperthreading or it is possible to render 2 files at the same time. You can imagine what you can do if you had hyperthreaded multicore processors.
That's not to say that software can't be written to take advantage of all types mentioned.
Single core have some advantages, especially in overclocking, where dual core finds that difficult as it shares comman internal architacture, so a single core (at this stage) is great for games, the best being the old AMD FX53 now sold as the 4000+
Multi CPU's of course work best with OS and software to support them.

Re:Multicore vs Single vs Hyperthread vs Multi CPU

[LubosD]

Normally a computer would wait or at best lag behind while it's looking for processing cycles.

Yeah, certain operating system(s) have big problems under high loads but the other don't ;-)

I want single core COLD processors!

[Otis_INF]

I don't need n, n>1, cores on a die, I want a processor which is COLD, so it can be passively cooled, or with a very small cooler. The fun of those things is that it will make computers silent again, like in the days of the 486 and earlier. Proper motherboard designs also can allow multi-proc systems which will mitigate the fact that you need multiple cores on a die.

Re:I want single core COLD processors!

[Anonymous Coward]

They had those after the 486. In fact, I've got two passively-cooled Pentium 2s. I suppose that's the one advantage of the CPU being so friggin' huge, more heatsink space.

Re:I want single core COLD processors!

[swordgeek]

Well for cold processing, you should look at Sun's new T1-based systems. Eight cores on a single die, four threads per core, and all in 72W maximum.

"Proper motherboard designs also can allow multi-proc systems which will mitigate the fact that you need multiple cores on a die."

The problem with that idea is that as soon as you have a multi-socket motherboard, you can put multi-core CPUs in each socket. In actual fact, multiple cores make a lot of sense over discrete processors, depending on the bus architecture, but multicore is where things are going to go.

As for silent computers, that is finally getting the attention it deserves, thanks primarily to the HTPC craze. I would love to see limited power CPUs with no active cooling systems in them.

Duh

[diegocgteleline.es]

Many of the design decisions taken in dragonflybsd are similar than those from linux, so I don't really see the design "magic" that doesn't allows linux to scale "without corporate support". Sure, dragonfly looks good (it still has a LOT of work to do) but is not that linux sucks in this field.

Linux was redesigned from the ground up to scale well to huge multiprocessor systems. This is why it has been able to go from being not 100% optimal with 4 cpus to scale properly to 512-cpu SGI machines in a single development release cycle.

It's not just corporate support what allowed this - many years ago people suggested that linux should fight scalability going the "solaris way", but linus & other people didn't like that way at all. Can you explain what radical design concepts are going to allow dragonfly to behave so much better than linux and/or solaris in 512-cpu machines? (yes, I know that dragonfly is also targetting to become a "clustered os")

Re:Duh

[Mancat]

Linux was redesigned from the ground up to scale well to huge multiprocessor systems.

LOL! If you say so.

Re:Duh

[diegocgteleline.es]

Maybe you missed the "re" in "redesign"? ;)

amd vs. intel (ftfa) and the mhz clusterf#ck

[sl4shd0rk]

"Apjit Walia, an analyst at RBC Capital Markets, agreed, noting that AMD is enjoying at least the perception that its products are better than Intel's chips."

Interesting quote. I've never seen an intel chip outperform an amd chip. Even the k62-450 had more snap to it over the PII-450. I replaced three dell boxes a couple months ago that had p3 3.4 processors in them with fx-55 homegrowns. The guys that use them say they never want their intel boxes back.

We ran a test on the 3.4 P3 and the fx-55, converting a 20 meg Parasolid into an IGES file (this is all cad work). The p3 finished in something like 3 mintes. The fx-55 did it 1.75 minutes.

The speed difference is 3.4ghz vs. 2.8ghz. you can see those numbers make no sense anymore.

News ? It was a little fresher in 2004....

[Anonymous Coward]

Gee whiz, doesn't anyone have a memory these days ?

This was all planned out YEARS ago.

http://www.theregister.co.uk/2004/08/16/amd_quad-c ore_opteron/

Re:Octacore - T1 "weak cores"?

[Anonymous Coward]

Huh?

A computer is measured on the ability to get real work done quickly, not what year the design was created. That is irrelevant. The 8 core T1 CPU supports 32 native threads running since most of the time a CPU is waiting for IO (ram, disk). If Sun got this so wrong, then why is Intel also following this technique?

The workload is important, so for a DB server, the T1 CPU isn't the best fit. But for an application server - stand back! Checkout the price/performance ratio.

Due to non-disclosures that I don't know if they've expired, I can't provide any more details. The 8 core T2000 will eat the lunch performing application server duties of a group of CPUs from Intel (or AMD) that add up to 8 cores for the same cost. See here: http://www.spec.org/.

I will give you that almost any current multicore CPU will toast the T1 on floating point - there's only 1 FPU shared across all 8 cores.

Last email of Randy Allen leaked

[MK_CSGuy]

"Mr. President, we must not allow a Quad-Core gap!!"

Re:The one thing I don't do, is video transcoding.

[lubricated]

>> Neither of these tasks is important for a desktop user. Large software compiles, scientific analysis -- these are specialized tasks and sure multiprocessor machines will help. The market for these, however, is specialized and limited.

You could have said this for the past 20 years and been correct. Somehow though most modern software would do poorly on a 486.

AMD and Intel agrree with you

[AHumbleOpinion]

Say bye to the race to the Gigahertz. Say hello to the race to the core count.

Well at last year's Game Developer Conference AMD and Intel were both telling attendees that future performance improvements will be coming more from multiple cores and less from clockrate. Both were hammering on the importnce of making your code multithreaded, both were advocating OpenMP.

Re:AMD and Intel agrree with you

[4D6963]

"future performance improvements will be coming more from multiple cores and less from clockrate"

Yup, as I said in another post, they're striving to get to higher clockrates while maintaining a decent power usage so now they really do need an alternative to keep going while lowering the power usage.

That's only partly true...

[CFD339]

Its really only very recently that bus speeds have been so far outstripped by processor speeds. If memory serves - as late as 1991, when I was still doing cube farm work, I was given an NEC desktop machine in which one key selling point was that it had a bus path to the IDE drive at the same speed as the process in mhz. Literally a 1:1 ratio. Today, a quad core 3ghz processor would be (in theory) handling something not quite equivilant to 12ghz, but talking to i/o on a bus of what, 800mhz or perhaps 1ghz if things progress as they have been? That's a 12:1 ratio.

Yes, I understand that multiplying the speeds like that is overly simplistic. An arguement could be made that its really more like 3:1 -- but since all the cores are using the same bus, I don't agree with that arguement.

Processors need to keep getting faster -- of course they do. Dual core is absolutely better than single core, and I'd never buy another new processor that wasn't dual core. I've enough experience on just hyperthreading chips to know for a fact that it "feels" smoother to operate the machine.

My point is only that this isn't the biggest factor any more. Until a better bus becomes popular and can handle the through put to support these supercomputers-on-a-chip, they won't continue to provide the kinds of "real world" performance increase that we'd expect.

Re:That's only partly true...

[moonbender]

It's not just the busses. Hard drives, RAM, none of it has kept up with the CPU speed increase. CPU/memory performance gap - old story, but very present problem.

Re:Octacore - T1 "weak cores"?

[Courageous]

CPU is waiting for IO (ram, disk).

My point exactly.

If Sun got this so wrong, then why is Intel also following this technique?

Because they are sell to people who don't know the point above.

C//

LOL, thanks for your concern.

[CFD339]

So far, the I've found the best way to avoid outsourcing is to continue to be creative and actually help people get things done. I've had my own business now for 15 years, and still have my first client. Outsourcing comes, and outsourcing goes -- and its not really the fault of the people in the sweatshops. When you outsource, you get exactly what you ask for. Nothing less, but nothing more -- and most clients don't know to ask for the right things, so it fails.

Is my code work boring? Well, I'm having fun with it -- and its saving lives & property. You?

Re:Octacore - T1 "weak cores"?

[Courageous]

See here: http://www.spec.org/ [spec.org].

T1000 and T2000 do not appear as a result of the search. Can you point me to the result?

C//

If you actually bothered to read the thread....

[CFD339]

You'd see the I specifically referenced video transcoding as an exception -- and that for people doing this a great deal, a faster bus would allow inexpensive dedicated processing products to handle this much more efficiently for the processor.

Re:DragonFly BSD will really start to shine.

[m.dillon]

Well, I wouldn't call DragonFly revolutionary yet, but a lot of people don't understand that our ultimate goal is closer to the idea of the network as the computer rather then the computer in the network.

In otherwords, what we are gunning for is the ability to create widely distributed computing sets not only within a LAN environment, but within a WAN environment ('the internet') as well, and for these sets of resources to operate transparently and appear as a single machine to the user.

To make this work pretty much the entire kernel has to be rewritten to properly incorporate the cache coherency management required to actually be able to operate efficiently in such an environment, and you aren't going to see much differentiation until we complete that process. Part of the work involves a true separation of work, which means threading operations (such as the TCP protocol stack) without having to rely on mutexes and other serialization methods that would otherwise require SMP/NUMA architectures to be efficient. The heart of the problem, though, is filesystem-level cache coherency and cache management. If a workload is operating across a slow WAN link with 50ms of latency you simply cannot afford to have data managed via synchronous protocols. The individual sites have to be able to manage data sets in a fully cache coherent manner with the only synchronous communication occuring only when there is a conflict requiring resolution. That isn't an easy problem to solve.

-Matt

Memory and CPU usage has changed too..

[Chordonblue]

MAC OS X is certainly not the only platform with increased requirements. About the time of the 200-300 MHz processor, the Windows OS of your average computer only used up 30 or so megabytes. Have you looked at the average computer TODAY at boot up?

See, not only do we have to run an anti-virus program in the background (with a HUGE list of viruses to compare against), but not we also appear to need anti-spyware and firewall programs as well. Then you have the additional overhead of things like 'Active Desktop' and, yeah, all of a sudden your 300 MHz AMD K-6 makes Word 'run slow'. Sure, it's not Word's fault, but the platform it's running on is bogged down by all this crap.

Excellent Analogy

[raftpeople]

Very good!

good and bad of multi-core

[goldfita]

Back when I was an undergrad, I wrote a paper on how asynchronous design would be a partial solution to the clock speed problem. Then this multicore thing seemed to come out of nowhere. I don't think it's all that much different (and I'm not familiar with the details) from having multiple cpus. It's less space and might be lower cost/heat/power. But it is also a good solution for many problems.

There are a number of posts that mention threaded code. But remember, it's not just threads, but processes. New services and apps are being introduced all the time. You may have things like ftp, ssh, email, bluetooth (in my case) running on your machine. I also tend to have a lot of applications open - browser, editor, and compiler just to start with. None of these are incredible cpu hogs, but if you're downloading a file while compiling and typing into text editor, you could use a little more power. Have you ever gone into a library and seen dozens of computers (or just terminals) in use or waiting to be used? Imagine if there was just one multicore machine (say 32) and all the terminals were working with that machine. That would save a lot of money, possibly space, and power. You might want to have one physical server hosting dozens of services. In these situations, no (or little) extra work is necessary to make use of the additional cores.

However, there are still those problems that need Hz. Many scientific/math problems are not parallelizeable. Unfortunately, multicore does nothing here. Oh well, I'm happy with my dual cpu machine. If only it would stop hanging...

As has been said....

[Anonymous Coward]

Dual core super duper pentium extremes with hyper diaper threading are only good if you like cracking 256 bit encrypted cheese rolls or for scientific processing or a similarly processor intensive task.

Most of all you need good I/O processing and then you dont have to worry about power, of course a fairly good CPU always helps anything.

So yeah while your focusing on your Penti yum! wonderbook extreme complete with enhanced chimes and wonderfully disney like american names running Windows 64 bit Powerhungry edition, I'll be enjoying my nice little book with the right specs and the right OS (for me) ... Windows 2000 FreeBSD and Ubuntu Linux (multi boot).

current AMD chips...

[YesIAmAScript]

Use two busses to talk to the RAM. They copied this from Intel (who didn't invent it themselves). This is the "two bank interleaving".

As to your comment of directly linking core with memory, that either isn't possible, or is already done, depending your particular definitions.

You connect two things with a bus, and SDRAM (including DDR) signalling isn't compatible with the normal addressing in a CPU, so you're always going to need to convert between the two. Given how SDRAM is internally designed, it doesn't make sense to change the interface to the SDRAM, it won't become any more efficient and it'll take more pins.

OT swearing

[David Rolfe]

Take that, Karma!

Me making a funny typo: Why you would need to double mod it I have know idea.

You, insultingly: "Know" idea indeed. This must be a troll. I have no idea how one could seriously try to be a pedant and fail so miserably. Most morons wouldn't even try; you must be a special kind of moron.

Lol! Let the ad hominems fly. You're mother is a corpse fucker, but it's OK because your dad can't feel it anyway. Yay! But yes, even by replying in the first place I could have easily been considered trolling (just like you!). Anyway, rather than addressing the question 'is nitpicking style even pedantry,' I guess we'll never know! Well, since this is our last meeting: Farewell. (I foed you, so you needn't bother replying :) I won't see it.)

Would you also use two unnecessary commas in your essay?

It is grammatical to use a comma before the conjunction and. A comma separating the independent clause is also grammatical. Look it up.