Intel's "Terascale" Vision

Good by betterunixthanunix · 2006-09-27 08:18 · Score: 3, Funny

Now I can run 80 instances of Doom at the same time. Nothing quite like heavy multitasking.

--
Palm trees and 8

Re:Good by networkBoy · 2006-09-27 08:21 · Score: 2, Insightful

Heh,
Not really.
This chip (as designed) would be one CISC CPU core and 80 Mini cores (kinda like Cell?)
Anyway, where this will be awesome is in rendering &&|| Cryptography, where the memory bandwith requirements is not as high as CPU compute requirements.

I personally hope these come out in a 4xPCIe expansion card:)
-nB

--
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
Re:Good by rts008 · 2006-09-27 08:32 · Score: 1

Heh, add Dual Quad SLI, then maybe I can get 600fps in Tux^H^H^H PPracer!

--
Down With Slashdot BETA!!! I've been around the corner and seen the oliphant; you can only abuse me from your perspecti
Re:Good by $RANDOMLUSER · 2006-09-27 08:48 · Score: 1

This is gonna be interesting. Can you imagine Intel touting Cell's no-out-of-order, no-cache architecture? They'd have to disavow the last twenty years marketing efforts. The mind reels...

--
No folly is more costly than the folly of intolerant idealism. - Winston Churchill
Re:Good by HoboMaster · 2006-09-27 09:20 · Score: 1

Wouldn't surprise me. Intel does that. Consumers have very short memories when it comes to marketing.

--
Remember kids, tin foil doesn't work, so use LeadHat.
Re:Good by Jartan · 2006-09-27 09:58 · Score: 1

"This is gonna be interesting. Can you imagine Intel touting Cell's no-out-of-order, no-cache architecture? They'd have to disavow the last twenty years marketing efforts. The mind reels..."

Yes because "Intel inside" is clearly not going to work anymore!

They are going to have to change it to Intel's Inside
Re:Good by treak007 · 2006-09-27 11:26 · Score: 1

especially when intel decides to reformat the consumer's life hard drive.

--
Klingon Software is not released, it escapes, inflicting terrible damage onto the enemy as it does
Re:Good by obeythefist · 2006-09-27 12:50 · Score: 1

Please. 2 cores, 4 cores, 8 cores, it will never take off until someone invents multithreaded porn.

--
I am government man, come from the government. The government has sent me. -- G.I.R.

well now it seems by Compaq_Hater · 2006-09-27 08:18 · Score: 3, Funny

we are on our way to L-Cars computers i can feel it.
CH

Re:well now it seems by garcia · 2006-09-27 09:24 · Score: 1

8, 80, whatever cores? I just want my hovercraft damn it, they promised us hovercrafts!

DUP alert by John+Jamieson · 2006-09-27 08:19 · Score: 1, Funny

Yes, remember the days when Intel used to accuse AMD of copying everything?

Now it looks that Intel has to take a page from the SUN Niagra roadmap for inspiration.

Ahh, the good old days.

We have a dupe! by Anonymous Coward · 2006-09-27 08:19 · Score: 1, Informative

dupe

doioioioioi

Re:We have a dupe! by tonsofpcs · 2006-09-27 09:05 · Score: 4, Funny

Maybe we'll get 80 copies of this article.

--
Video Production Support

80 Submissions by nycsubway · 2006-09-27 08:19 · Score: 3, Funny

This processor must already be submitting stories... If it is there should be 78 more dupes just like it.

I like the idea of an 80 core processor. Multithreaded applications will work better. Why are people afraid of multiprocessors? Systems with dozens of processors are not uncommon. I dont see why it would be bad for the desktop.

--
http://github.com/gbook/nidb

Re:80 Submissions by Compaq_Hater · 2006-09-27 08:22 · Score: 1

It is usually cost that they are afraid of :), i know i would be. What do you think this thing will cost ?. I am going to guess about 5k maybe more who knows.
CH
Re:80 Submissions by Goblez · 2006-09-27 08:28 · Score: 1

How about due to the lack of code that takes advantage of the multiple processors? If you mainly use one heavy application that doesn't take advatage of more than one or two cores, then those other 78 are going to be bored (and not submitting their dupes to ./)!

--
- Kal`Goblez
Re:80 Submissions by nycsubway · 2006-09-27 08:34 · Score: 4, Insightful

That is true. A lot of applications do not heavily use multithreading. But, in the scientific community a lot of applications require it. Where I work, we process several GB of MRI data a day. We are able to parallelize the overall processing, so the more processors, the better. However, I wish Matlab would become multithreaded! Our servers have 4 processors and if matlab used them all, we could process 1 dataset in 1/4 the time, instead of processing 4 datasets at once to utilize the CPUs. Processing one dataset at a time would reduce disk I/O.

--
http://github.com/gbook/nidb
Re:80 Submissions by NerveGas · 2006-09-27 08:36 · Score: 2, Insightful

It's not the number of processors or cores that they're afraid of, it's the fact that with the exception of a very few cases, your performance does not scale linearly with the number of CPUS, it is less than 1:1. To make it worse, as the number of CPUs rises, the cost to intelligently, quickly deliver sufficient bits to and from all of the CPUs gets exponentially higher.

Recently, some of our managers wanted to see what it would cost to purchase a system that would significantly outperform our 8-way Opteron for RDBMS work. I got numbers on machines from various manufacturers, and when the managers saw them, the conversation was instantly over.

steve

--
Oh, you're not stuck, you're just unable to let go of the onion rings.
Re:80 Submissions by zippthorne · 2006-09-27 08:47 · Score: 1

I could've sworn I saw something about a version of matlab compliled with mpich a while back, but I'll admit it could've just been a way to run multiple instances of it at once that i saw. If your situation does not require a whole lot of fiddling, you could put a lot of the code into a compiled .mex file and make that parallel.

Ironically, the nature of MATrix LABoratory's design goals is particularly suitable for multi-processor implimentations. The language is expressly designed to allow/coax users into thinking about problems in massively-parallel ways. It should've been one of the first programs ever to take such advantage.

--
Can you be Even More Awesome?!
Re:80 Submissions by Anonymous Coward · 2006-09-27 08:52 · Score: 0

There are a few 'parallel' MATLAB projects - one that pops immediately into mind is a product by "Interactive Supercomputing". Take a look at: http://www.interactivesupercomputing.com/

One of the guys involved is Alan Edelman -- he's a good guy and very knowledgeable. I have no ties to the company and have only met him in passing, but if it helps, hey, go for it! :)

Good luck,
- Anonymous
Re:80 Submissions by kabocox · 2006-09-27 09:09 · Score: 2, Interesting

That is true. A lot of applications do not heavily use multithreading. But, in the scientific community a lot of applications require it.

If anything, this will be the one great thing to come out of 8+ core desktop systems. I honestly don't think "most" apps even pretend to use more than 1 core very well. Once 8+ cores are one your bare bones Dell home PC then I'd expect to see everything under the sun start to be multithreaded. With the expectation of 32+ or 64+ cores in a decade time, then I could see alot of those little downloadable apps actually being designed to run in the background on just any given core. I'm actually eager to see some of the programming shifts that would take place because of this. From what I've read mostly /. is concerned about the memory bus not beening wide or fast enough to feed an 80 core chip. I'm more interested in "background" processes or algorithms designed just to load and run back there and only occasionally talk with other chips. I'm not worried that we can't figure it out. I'm kinda disappointed that we'd have to have 80+ cores out for a decade or so before we really even start to scratch the surface of what we can do with them. Right now, everyone is thinking of hey you can't breakevery problem into parallel tasks. Well, what if we make an algorithm breakthrough that says we can? But we only find that out after having all that processing at hand? The good news is that all those supercomputer folks have been researching into this area for years. The bad part is that even they'll be surprised when your Joe Average CS person suddenly has access to that type of system. We don't know, yet what it'll really excel at. We'll figure it out after it is sold to 5% of the US public. ;)
Re:80 Submissions by dvice_null · 2006-09-27 09:17 · Score: 1

It is not uncommon that people run 30-50 processes on single workstation. Majority of these processes are usually related to operating system or other invisible tasks, but usually at least a few are visible processes, like web browser, antivirus, firewall, music/video player, IM, games, etc. Currently I don't think that there is a need for 80 cores, but even now there is work for many of them. And once we get the hardware, people will start writing software for it. And of course the rest of the cores could do some scientific research for boinc projects to benefit the human kind.

The reason why they are increasing the amount of cores is that nowadays they can add so many transistors on a single chip that they actually don't know what to do with them. At the same time people are using more and more programs at the same time on a single computer. So the obvious solution was to add more cores.

Also, this would propably be very usefull on cheap server.
Re:80 Submissions by m0nstr42 · 2006-09-27 09:26 · Score: 3, Interesting

How about due to the lack of code that takes advantage of the multiple processors? If you mainly use one heavy application that doesn't take advatage of more than one or two cores, then those other 78 are going to be bored (and not submitting their dupes to ./)!
I'm curious - supposing that the software existed to take advantage of it, would it be possible to design an operating system that used a vast number of cores in a radically different (and advantageous) way than we use one or two (or a few more) today? i.e. the kernel spawns several sub-kernels on different processors or clusters of processors, with each one set up to handle a very specific task. Is there really any advantage? In nature, large scale systems of simple agents tend to be able to accomplish complex tasks more efficiently than single agents or small groups.

--
m0nstr42.blogspot.com
Re:80 Submissions by Ironsides · 2006-09-27 09:27 · Score: 1

However, I wish Matlab would become multithreaded! Our servers have 4 processors and if matlab used them all, we could process 1 dataset in 1/4 the time, instead of processing 4 datasets at once to utilize the CPUs. Processing one dataset at a time would reduce disk I/O.

It's sounds like you are using a fairly high end computer system. I have to sugestions for upgrades to it that would help reduce disk IO times and increase the speed.

1) RAM Drive
2) Solid State Hard Drives

Depending on your system and data size would depend on which you get. However, I have seen a "cheep" solid state disk that was basically a compact flash card hooked into a compact flash to IDE adapter. Ram drives are all software and are probably faster. However, you would need more RAM for it.

--
Fly me to the moon Let me sing among those stars Let me see what spring is like On jupiter and mars
Re:80 Submissions by Doctor+Memory · 2006-09-27 10:25 · Score: 1

Right now, everyone is thinking of hey you can't break every problem into parallel tasks. Well, what if we make an algorithm breakthrough that says we can?
Good luck, I believe that particular problem (generalization, when some parallelism can be extracted within tasks) has been shown to be NP-hard. Even if you come up with a golly-gosh-whizzbang algorithm or pattern that takes maximum advantage of a multi-cpu system, you're still going to have a hard time forcing problems to fit it. My guess is that even when we've got 4-core processors on almost every desk, 80% of the software people use day-to-day will still block on a single CPU, just because the problem either isn't easily broken down, or not enough of it is (what good is it to have the initialization routine run in O(1/n), if the main body of the program doesn't?).

I think our best bet is to hope (or help) the major OS suppliers come up with some good general-purpose multi-tasking libraries, so at least some common operations will be able to take advantage of multiple CPUs.

--
Just junk food for thought...
Re:80 Submissions by archen · 2006-09-27 10:30 · Score: 2, Informative

I think that was part of what the article was inferring. Assuming you had a perfectly optimized kernel and a zillion cores, performance still isn't going to scale all that well. There is just too many bottlenecks all over the way the general purpose PC is designed today. And lets not forget how far behind hard drive tech is dragging compaired to the rest of the system. It's funny because everyone acts like this is so new despite the fact that high end stuff like supercomputers have been dealing with these issues for decades. The PC arcetecture is going to have to change in more than a couple ways, but before that happens everyone is going to have to get used to the fact that their system has more than one core. Maybe that's one of the tricks AMD has up its sleeves. Buying ATI may have been a step towards re-engineering the parts of the PC that are going to be bottlenecks.
Re:80 Submissions by powerlord · 2006-09-27 10:35 · Score: 1

I wonder if something like GNU's Hurd where "The Hurd is a collection of servers that run on the Mach microkernel to implement file systems, network protocols, file access control, and other features that are implemented by the Unix kernel or similar kernels (such as Linux)." might benefit from the a massively multi-core environment?

Granted it still needs a lot of work, but it might provide a faster system then the monolithic kernels of Linux and Windows.

--
This space for rent. All reasonable inquiries will be entertained at proprietors discretion.
Re:80 Submissions by Anonymous Coward · 2006-09-27 10:42 · Score: 0

Feel relieved; Mathworks is working on this problem. I attended a talk at SIAM's conference on Parallel Processing for Scientific Computing in February. Cleve Moler (and he would know) said that Mathworks currently has a large initiative in parallel computing. As you might imagine, everyone there wanted to know when the technology would be delivered. However, he refused to put a date on a release and simply commented that it would be released when they had done it right.
Re:80 Submissions by JourneyExpertApe · 2006-09-27 11:28 · Score: 1

Matlab's good for testing and small calculations, but if you want to do some serious computing, you should consider learning C++ and writing the code yourself. Then, you can create as many threads as you want.

--
If you can read this sig, you're too close.
Re:80 Submissions by rufty_tufty · 2006-09-27 11:50 · Score: 1

Supercomputers are devices for turning Compute bound problems into I/O bound problems?
if every desktop is bound by I/O, then that's an interesting place to be.

--
"The weirdest thing about a mind, is that every answer that you find, is the basis of a brand new cliche" -
Re:80 Submissions by alphamugwump · 2006-09-27 12:55 · Score: 1

Well, most people don't use all of their single core cycles either. How many people do you know who use their "extreme gaming rigs" mostly for web surfing? Eh?

Nah, with the added horsepower, applications will grow to fit the hardware. A few people will actually need and use those extra cores, probably for things like data crunching, image/sound processing, and AI.
Re:80 Submissions by TheViewFromTheGround · 2006-09-27 13:12 · Score: 1

Another question along these lines is whether it is even remotely cost effective to build general-purpose OS architectures that can handle your proposition, and how much overhead is required for your OS/meta-OS to supervise the whole affair. I've got a buddy who does some of this kind of work at a big research lab, but they are not writing a general OS kernel; they've been writing software to model a teeny sliver of quantum-scale physical phenomena, and they've been working on it for 10-15 years.

My hunch is that the only way we'll see an OS that could really intelligently and creatively do what you propose on the desktop is if it comes out of academic or defense research in pursuit of some other goal.

--
Online citizen journalism from the inner city: The View From The Ground
Re:80 Submissions by Alioth · 2006-09-27 21:38 · Score: 1

But when it first does, expect newly multithreaded applications to be very buggy and unstable. Writing multithreaded applications that work well is a technique often not understood by developers who up until now have been doing everything singlethreadedly. Race conditions and deadlocks will abound, and I'm sure there will be new classes of security exploits.

--
Oolite: Elite-like game. For Mac, Linux and Windows

Time to go home... by Aqua_boy17 · 2006-09-27 08:19 · Score: 4, Funny

Anyone else first read that as "Intel's Testicle Vision"?

Man, it's been a long day.

--
What if the Hokey Pokey really is what it's all about?

Re:Time to go home... by elsilver · 2006-09-27 08:39 · Score: 1

Nope. I got "Intel's Tentacle Vision".
Either way, gives new meaning to "pervasive computing". (Or is that "invasive"?)
E.
Re:Time to go home... by Apocalypse111 · 2006-09-27 08:40 · Score: 1

My first read was "Intel's "Tentacle" Vision".

I'm a sick, sick man.

--
There is no mod option "-1: Disagree" for a reason. "Overrated" is not an acceptable substitute. Post something instead.
Re:Time to go home... by Anonymous Coward · 2006-09-27 08:44 · Score: 0

I'm afraid I did too....

Now I need to double check I am posting anonymously.
Re:Time to go home... by jacobw · 2006-09-27 08:45 · Score: 1

Anyone else first read that as "Intel's Testicle Vision"?

Hmm... I guess multiple cores would be useful there, too, but everybody would need bigger boxer shorts. Plus, the cooling requirements would be steep. Nothing kills the moment like the roar of a fan kicking into action...

--
Arr! Read The Government Manual for New Pirates!
Re:Time to go home... by markana · 2006-09-27 09:36 · Score: 3, Funny

Then they would have to name it the "Hentaium" processor...
Re:Time to go home... by KarmaMB84 · 2006-09-27 09:42 · Score: 1

I thought it said "Tentical". I thought perhaps Intel was going to build Japanese schoolgirl molesting products for a second.
Re:Time to go home... by cxreg · 2006-09-27 09:49 · Score: 1

Anyone else first read that as "Intel's Testicle Vision"?

No. Nobody else thought that. You are a pervert and a bad, bad person.

--

God Fucking Damnit
Re:Time to go home... by kurzweilfreak · 2006-09-27 10:36 · Score: 1

You win the "best post of the day" award. Thank you. :)

--
kurzweil_freak
5th Kyu Genbukan Ninpo/KJJR student
Be the darkness that allows the light to shine.
Re:Time to go home... by dp_wiz · 2006-09-27 19:11 · Score: 0

"perversive computing"

I see by Anonymous Coward · 2006-09-27 08:20 · Score: 0

More Cores = More Dupes?

Why have 8 strong ox? by chroot_james · 2006-09-27 08:24 · Score: 4, Interesting

When you can have 80 underfed chickens?

--
Reality is nothing but a collective hunch.

Re:Why have 8 strong ox? by Compaq_Hater · 2006-09-27 08:27 · Score: 2, Funny

Hell yeah, now Micro$oft can write Windows BBQ edition that will fix those OX and Chickens.
CH
Re:Why have 8 strong ox? by Anonymous Coward · 2006-09-27 08:36 · Score: 1, Informative

FYI, the plural of ox is oxen.
Re:Why have 8 strong ox? by Pharmboy · 2006-09-27 08:49 · Score: 1

Actually, I am guessing that someone at Intel has been taking the "imagine a beowulf of those" jokes on slashdot too serious, and decided to put it on one chip.

Tho is *seems* that if the OS was written specifically for 80 cores (ie: 64 bit, one bit per core or something) then *if* they synced up nicely, you could do some cool stuff with games at the least. My guess is that getting the OS to work with 80 cores in near real time is going to produce some serious overhead, however. For what it is worth, $10 says it will run Linux before it runs Windows, but that prediction isn't exactly going out on a limb.

--
Tequila: It's not just for breakfast anymore!
Re:Why have 8 strong ox? by WilliamSChips · 2006-09-27 08:57 · Score: 2, Insightful

Because the chickens actually work better now and most oxen are now really just three chickens yoked together--hell, most chickens are three chickens yoked together!

--
Please, for the good of Humanity, vote Obama.
Re:Why have 8 strong ox? by Yokaze · 2006-09-27 09:11 · Score: 1

Because you can't breed oxen from a pair of chicken and you can only put maybe a single ox in a hen-coop for 80 chickens? Figuratively speaking.

--
"Between strong and weak, between rich and poor [...], it is freedom which oppresses and the law which sets free"
Re:Why have 8 strong ox? by Anonymous Coward · 2006-09-27 09:15 · Score: 0

Underfed? This technology demonstration has sufficient bandwidth to feed data to (and sink data from) all of the 80 simple cores without issue.

On specific tasks this little 80 core device would eat 8 oxen for dinner and then some.
Re:Why have 8 strong ox? by walt-sjc · 2006-09-27 09:18 · Score: 1

And 80 cores worth of blistering heat to cook over... Yee haw!
Re:Why have 8 strong ox? by RingDev · 2006-09-27 09:21 · Score: 1

But in this case, those 80 underfed chickens will have lasers strapped to their heads.

-Rick

--
"Most people in the U.S. wouldn't know they live in a tyrannical state if it walked up and grabbed their junk." - MyFirs
Re:Why have 8 strong ox? by mikael · 2006-09-27 09:24 · Score: 1

If it is possible to get Linux or Windows to run on two core system, it shouldn't be too much trouble to run on a 64 or 80 core system. Although, I'd be worried about getting 80+ shell terminal messages:

CPU0: Temperature above threshold
CPU0: Running in modulated clock mode ...
CPU80: Temperature above threshold
CPU80: Running in modulated clock mode

--
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Re:Why have 8 strong ox? by HoboMaster · 2006-09-27 09:28 · Score: 3, Funny

"So, Brian, what's the plural of box?"
"Boxen?"

Gotta love Brian Regen.

--
Remember kids, tin foil doesn't work, so use LeadHat.
Re:Why have 8 strong ox? by Anonymous Coward · 2006-09-27 09:57 · Score: 0

You tell 'em. That'll fix their ox-cart.
Re:Why have 8 strong ox? by fotbr · 2006-09-27 10:04 · Score: 1

So...imagine a beowulf cluster of THOSE.

Actually, as far as uses for it go, I think games will be pretty low on the list. Scientific computing will probably be far and away the number one use for it -- IBM will probably come up with a new supercomputer they can sell based on racks of those instead of racks of opterons, etc.

I'd also guess that NSA type agencies around the world would like them as well - X years of CPU time doesn't seem like much when you've got 80 of them in one box, and you've got acres of those boxes -- although cryptography and data mining arguably fall under scientific computing even when they're not being used in pursuit of science. Then again, if they've already got something better, maybe they don't care about 80 intel processors per chip.
Re:Why have 8 strong ox? by Jesterboy · 2006-09-27 10:49 · Score: 1

The 80 chickens could ride around on the 8 oxen; now that would be fast.

And if it ever overheated? Mmm, tasty...
Re:Why have 8 strong ox? by Pharmboy · 2006-09-27 10:59 · Score: 1

Let me rephrase that then: running *well*. Not just booting, but actually efficiently using the 80 cpus.

--
Tequila: It's not just for breakfast anymore!
Re:Why have 8 strong ox? by Pharmboy · 2006-09-27 11:02 · Score: 1

But if Intel wants to sell any reasonable QUANTITY of them, which is necessary to get the scales of economy working for them, they need to appeal to a broader market. Porn and games are about as broad as it gets, I'm just sure why you need 80 cores to watch porn.

--
Tequila: It's not just for breakfast anymore!
Re:Why have 8 strong ox? by rufty_tufty · 2006-09-27 11:47 · Score: 1

I guess you were going for funny here, but:
Why have 8 Oxen doing speculative operations (and failing most of the time) when you could have 80 chickens doing only what you needed.
i.e if you accept memory badwidth is your limit, then only do the instrucitons you have to. 200 processes across 80 cores doing only the things they have to vs a few cores doing branch prediction and guessin like hell, the chickens sounds like a better deal to me.

--
"The weirdest thing about a mind, is that every answer that you find, is the basis of a brand new cliche" -
Re:Why have 8 strong ox? by fotbr · 2006-09-27 16:02 · Score: 1

Ah, but you forget....if your buyer is paying out of the "black" portion of the budget, economies of scale do not necessarily apply.
Re:Why have 8 strong ox? by Compaq_Hater · 2006-09-27 20:05 · Score: 1

Yes, I am aware of this and in true /. style refuse to use correct grammer or spelling :), just like Linux it is a matter of choice. But thanks for the Info :).

CH

Re:cock by Anonymous Coward · 2006-09-27 08:25 · Score: 0

That's what Zeus sucks!

imagine a beowulf... by SpaceballsTheUserNam · 2006-09-27 08:28 · Score: 0

i always thought that beowulf was a great story? once you get past the prose.

--
\.

In other news by Hahnsoo · 2006-09-27 08:29 · Score: 4, Funny

Gilette is releasing a new shaver called the "Plutonium Mach80", a razor with 80 blades. Each blade has a separate distinct function, and you can get even closer shaves with the synergistic cuisinart action. Also comes in a "For Women" model for "sensitive areas". "Basically, 5 blades isn't enough. I mean, really, more is better, right?", says Gilette CEO James Kilts. Schick is reportedly working on a competitor blade that may exceed the legendary "100 blade barrier".

Let's go with by Anonymous Coward · 2006-09-27 08:30 · Score: 0

160-Cores that way I'd have Tera-Terascale Server

Re:Let's go with by psbrogna · 2006-09-27 09:09 · Score: 1

You need a refresher on units. 160 cores would be a Duo Terra.
It's disconcertingto me how much Latin and general linquistics I am now burdened with to understand contemporary branding. How long before we start seeing IMG SOURCE="icon128.gif" ALT="The Ancient Rune Previously Known as Intel" ?

80 cores... by windowpain · 2006-09-27 08:31 · Score: 3, Funny

And slashdotters will still be overclocking the sumbitch.

--
Insert witty sig here.

Multicore processors are here to stay!! by Aditi.Tuteja · 2006-09-27 08:31 · Score: 0

Thinking about of multicore processors. Its performance is the first thing to ponder about...Imagine how it will ease our lives...while one processor core is updating the terminal display, another processor core could be tasked with processing the user input. As servers they can handle better backend processing. For example where there are transactions arriving from the various Point-Of-Sale terminals. By taking advantage of multi-core processor, any one server could handle a greater number of transactions in the desired response time. Its Gonna be real Fast and efficient..:)

Re:Multicore processors are here to stay!! by vadim_t · 2006-09-27 08:43 · Score: 1

We can already do that, on single core CPUs. It's called "multitasking", and things like that are usually done by using multiple threads or multiple processes. It's nothing that can't be done already, but coding threads is a huge pain.

Besides, 1 core would be better than several in many respects. Say, if you've got a web server, or a database or whatever this means that each process/thread only can get up to 1/80 of the processor's capabilities, unless the program is coded in such a way it can take advantage of multiple cores in a single operation (and even if it can do that, chances are it can only split the task into a few threads, so it still can't use the whole CPU)

Now, for systems with a huge load of small tasks, like a web server serving a really massive number of clients this will be really nice. But this kind of thing would really suck for a desktop. They rarely have more than a couple processes running at once.
Re:Multicore processors are here to stay!! by Ant+P. · 2006-09-27 11:23 · Score: 1

This thing would be great in a laptop: if you're only running 5 processes or so at once, you can turn off nearly all of the CPU.
Re:Multicore processors are here to stay!! by vadim_t · 2006-09-27 11:52 · Score: 1

You can do that already as well.

My Pentium M can slow down from 1600 MHz to 600, and then there are additional throttling states that seem to be able to bring it down to the performance of a 386. I made mine do that when battery got low, and at the highest levels of slowdown you could easily notice that even routine things like scrolling text in a text editor started becoming slow.

For reference, I'm typing this on a dual Athlon MP 2000+. That means I've got two CPUs available. This is under Linux 112 processes. I've got Second Life taking one CPU, and the load average is 1.42. Meaning, I'm not even effectively using the two CPUs I have. 1.0 of that would be SL, and the 0.42 is a combination of X, browser playing animations, and other stuff like that. A load average of 5 very ocassionally happens on my server, but is also uncommon.

I repeat: With current software, pretty much anything above a dual core will give you about zilch benefit unless you do something uncommon and parallelizable (rendering, encoding, compiling, etc). It's of no use for games, and absolutely useless for common desktop usage (unless you were going to open 80 browser tabs at once).

I don't doubt this will be a neat thing for large webservers, but it's even counterproductive for 99% of normal people.
Re:Multicore processors are here to stay!! by Ant+P. · 2006-09-28 03:10 · Score: 1

Load averages aren't just CPU usage. My non-HT P4 can get up to 7.5 on some days.
Re:Multicore processors are here to stay!! by vadim_t · 2006-09-28 04:03 · Score: 1

Load average is all about CPU usage. It's the average number of processes that wanted CPU time during the last N minutes.

So, your load average means that sometimes you could use a 8-way box. Take your load average, round up to the next integer, and that's the number of CPUs your workload could use. If you have less CPUs than that, then you're not getting all the performance possible, if you have more of them then they're going to waste.

Didn't sun try this? by NerveGas · 2006-09-27 08:32 · Score: 1

Didn't Sun try that sort of idea with the UltraSparc T1? If I recall correctly, while the concept of lots of light cores was cool, the real-world performance didn't do any better than Intel- or AMD-based systems.

steve

--
Oh, you're not stuck, you're just unable to let go of the onion rings.

Re:Didn't sun try this? by psbrogna · 2006-09-27 08:54 · Score: 1

If my conversations with "average" purchasers of hardware are any indication, real world performance is not a closely examined metric. How many of us know somebody that spent a metric butt ton of cash on a Duo laptop for browsing the web and word processing? I wish I had a block diagram (like the old old von Neumann architecture picture from early comp sci text books) that shows how insanely out of wack the interconnecting lines have gotten between modern computer sub components and especially including the dotted line to the internet cloud.

Intel's "Terascale" Vision by CaptKilljoy · 2006-09-27 08:35 · Score: 5, Funny

If they succeed, does this meen the tera-rists have won?

While 80 cores is pretty ridiculous... by foxtrot · 2006-09-27 08:35 · Score: 4, Insightful

A lab prototype like this can help them with something important: Given multi-core processors look to be the way future computers will be built, how do you feed them data? The current paradigm won't scale past 4 cores on a single chip's worth of FSB, and there are folks who don't think that even 4's going to be a useful increase over 2.

Even if Intel never sells a chip bigger than 16 or 32 ways, an 80 core lab mule will teach them many things about how to get information to a processor and keep those caches full of appropriate data.

-F

Re:While 80 cores is pretty ridiculous... by BSonline · 2006-09-27 09:59 · Score: 1

They've already addressed the FSB issues. They are going to use fiber optics for a LOT of the data transfer around the processor, including the FSB.

--
PS: That is what part of the alphabet would look like if the letters "Q" and "R" were removed.
Re:While 80 cores is pretty ridiculous... by Joe+The+Dragon · 2006-09-27 10:11 · Score: 1

how meny FSB links is it have?
Re:While 80 cores is pretty ridiculous... by BSonline · 2006-09-27 10:20 · Score: 1

I is not sure how meny is be. All 80 cores are linked, and the bandwidth available to them is in the TF range. The optical links go out from the processor, if I remember correctly. So they can run as many parallel links as they like...

--
PS: That is what part of the alphabet would look like if the letters "Q" and "R" were removed.

Make each core specialized!! by Anonymous Coward · 2006-09-27 08:36 · Score: 1, Interesting

If you are going to have 80 cores on a chip ..specialize some of the cores. Ie, have a few physics processors, a couple of graphics processors, maybe one dedicated to search tree retrievals even. One or two that are highly media centric. So basically 64 cores for general purpose CPU and the other 16 for known common tasks such as graphics and hdtv/media processing.
This would eliminate the need for a separate graphics card for the average business person and bring down costs.

Then, 80 cores should be enough for most people(TM).

For home PC's another option would be to have a Gigabyte of L2 cache, so that you are able to store the entire operating system image right next to the CPU. That would probably speed things up more than 8 cores would.

Re:Make each core specialized!! by NerveGas · 2006-09-27 08:39 · Score: 4, Insightful

Your comments are a lot more true than many people realize. Specialized hardware always wins.

As an example, people talk about using using multi-GHz machines for TIVO-type appliances, and "getting away" with 600 Mhz or so if your card has hardware MPG encoding. Some of the original TIVOs, because of their reliance on specialized chips and ASICs, used measly 33 MHz CPUs - and worked just fine.

steve

--
Oh, you're not stuck, you're just unable to let go of the onion rings.
Re:Make each core specialized!! by jawtheshark · 2006-09-27 08:46 · Score: 1

One word: Amiga... :-)

--
Ahhh...the great dumpster continuum. Many a free computer will be found there. -- sowth (748135)
Re:Make each core specialized!! by TheSHAD0W · 2006-09-27 08:47 · Score: 1

I would also take some of the memory used for cache on each core and make it directly addressable instead. This would allow the CPUs to operate more independently and efficiently on certain tasks.
Re:Make each core specialized!! by tppublic · 2006-09-27 08:59 · Score: 1

Your comments are a lot more true than many people realize. Specialized hardware always wins.
Define "win".
Seriously, while I agree that accelerated hardware can outperform specialized hardware from a performance perspective, that is not all there is to "winning". Ask Sony about Betamax.
Specialized cores may be useful, but it will only be useful in some (high volume or high value add) applications. Take networking equipment. If you strip down a core router (high value add), I'm sure the vast majority of the silicon is custom logic, mostly ASICs. However, if you strip down a branch router (where the volumes are 100X+ that of core routers, but still not high), you will find commodity silicon and differentiation in software. It's cell phones and other consumer devices where one is back to having custom accelerated hardware, because the volumes justify the savings that can be achieved over millions of units.
There is also a point to be made for software defined radios. While specialization may produce small decoding cores, the flexibility of performing decoding in software (and being able to do processing for many encoding schemes) may reduce the number of cores required to the point where a faster general purpose machine will outperform (from a total cost perspective) a specialized solution. It also has the potential advantage of being field upgradeable, something dedicated hardware cannot do.
Winning is relative, as there are many viable markets where hardware acceleration loses due to business realities.
Re:Make each core specialized!! by Anonymous Coward · 2006-09-27 08:59 · Score: 0

Oh the irony.. Can't locate the post from yesterday at the moment, but I just saw a post on yesterday's thread stating that general purpose always wins. Sounds like it's time for someone to weigh in on the facts to try and come up with a solution.
Re:Make each core specialized!! by dfghjk · 2006-09-27 10:21 · Score: 1

Funny, I'd say specialized hardware always loses.

Adding dedicated hardware for common tasks is not unusual, but in the long run it's commonly replaced with software. General purpose computing expands to consume all tasks as it gets faster. The best use of available die size is another blazingly fast CPU core, not specialized logic for specialized tasks that most people will not use.

There's a reason for the "S" in "SMP". The job of programming them is much easier.
Re:Make each core specialized!! by Sebastopol · 2006-09-27 10:39 · Score: 1

> Some of the original TIVOs, because of their reliance on specialized chips and ASICs, used measly 33 MHz CPUs - and worked just fine.

The answer is in this comment.

If 33 MHz "worked just fine" and TiVo upgraded anyway, why? Obviously they aren't under pressure from Intel, so the only answer can be "because 33 MHz CPUs did NOT work just fine". They wanted more content, more flexibility, more expandability, and needed a general purpose CPU. Theoretically you can make an ASIC for anything, but a general purpose programmable CPU is far, far easier to work with.

--
https://www.accountkiller.com/removal-requested
Re:Make each core specialized!! by Anonymous Coward · 2006-09-27 10:43 · Score: 0

why are nvidia OR ATI cards popular? Why is it that you never hear of software graphics acceleration anymore? And why not just use a second Celeron CPU instead of a graphics card? Oh yeah that's right there's lots of logic that a graphics card can do that would require multiple instruction cycles on a general purpose CPU (unless you increase the die area .. even then there may be a translation step/clock slowdown needed)

You said: "General purpose computing expands to consume all tasks as it gets faster."

Yes, this is true. Think about all they had to add to x86? Heard of MMX, SSE, SSE2 etc? That's what how they "expanded". That's all that crap multiplied by 80. The die area for MMX, SSE, and SSE2 repeated 80 times! This all increases the per core die area size. So you are telling me you rather increase the per core area?? This is an even BIGGER waste for features that "most people will not use".

Btw, the S in SMP is for symmetric. My one cpu desktop is already idle most of the time .. you're telling me that I'm going to need all 80 for desktop use? No, I'd rather save money and not have to spend it on a specialized graphics card if it's already on the CPU's die.
Re:Make each core specialized!! by NerveGas · 2006-09-27 15:39 · Score: 1

General-purpose often wins in value - because of the economies of scale. Specialized hardware always wins in performance. In markets where the "specialized" hardware can be manufactured and sold in quantity, then it wins in performance and is still price-competetive.

Look at a $50 video card. You simply could *not* produce 3D graphics like that from a general-purpose CPU (of any cost) at the same speed.

Yesterday's discussion was about commodity hardware as routers. They win in value until you hit the eventual limitations of general-puprpose hardware, but at that point, then it's like video cards: WIthout specialized hardware, you're just not going to cut it.

"Back in the day", even JPG compression/decompression was considered sufficiently heavy that there was a niche market in ASICs to handle it.

--
Oh, you're not stuck, you're just unable to let go of the onion rings.
Re:Make each core specialized!! by NerveGas · 2006-09-27 15:46 · Score: 1

Show me where video cards have been replaced with software. General-purpose isn't ever going to take over that. Physics is another area where general-purpose CPUs (even multi-core) don't have what it takes. (Whether you *need* that level of physics is another matter).

If a general-purpose CPU can handle your task, it will come out much cheaper. But there are still tasks where general-purpose CPUs just can't cut it.

Look at video encoding and decoding - a $20 ASIC can compete with a $300 CPU in terms of performance, while being produced on a massively inferior technology. Put a few of those ASICs on a 65nm CPU running at multi-GHz, and for a very small amount of real estate, you'd be able to encode dozens (hundreds?) of video streams in real-time. Again, whether enough people need that is another matter.

The only place that specialized hardware loses is in cost.

--
Oh, you're not stuck, you're just unable to let go of the onion rings.
Re:Make each core specialized!! by NerveGas · 2006-09-27 15:53 · Score: 1

I'm not entirely positive that they *did* upgrade. I haven't followed the hardware in a while, but it's entirely possible that some models are still running on those 33 MHz CPUs.

If they did upgrade, one reason would simply be that whatever chips are "commodity" at the moment are cheaper. Sort of like DIMMs, it's cheaper to buy 400 MHz DDR memory than plain old PC133.

steve

--
Oh, you're not stuck, you're just unable to let go of the onion rings.
Re:Make each core specialized!! by dfghjk · 2006-09-27 16:34 · Score: 1

"Why is it that you never hear of software graphics acceleration anymore?"

Actually I never heard of them. Software is what you have when you don't have hardware acceleration.

"And why not just use a second Celeron CPU instead of a graphics card?"

You could if you didn't want 3D.

"Think about all they had to add to x86? Heard of MMX, SSE, SSE2 etc?"

Yeah, I've heard of them. Better to add SIMD instructions to the CPU than to add a dedicated DSP. That would be the general purpose approach rather than the specific function approach. Thanks for proving my point.

"you're telling me that I'm going to need all 80 for desktop use?"

No, I never said that. Thanks for totally failing to understand. Enjoy your 80 specialized cores, no two of which are the same.
Re:Make each core specialized!! by dfghjk · 2006-09-27 16:44 · Score: 1

I never said that GPU's would get replaced, although dumb frame buffers are totally adequate for some applications. Not all computers need GPU's either so it would be stupid to integrate them as one of the cores in the main processor. Physics isn't something that many people need either.

"If a general-purpose CPU can handle your task, it will come out much cheaper. But there are still tasks where general-purpose CPUs just can't cut it."

Absolutely, and as time goes on fewer and fewer tasks can't be met.

"The only place that specialized hardware loses is in cost."

Let's see you run Word on specialized hardware.

Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-27 08:36 · Score: 1, Interesting

With 80 friggin cores, you'd darn well better get the architecture right. But that's unlikely to happen beyond the memory subsystem.

What REALLY needs to happen is to rearchitect the IO subsystems. Ala the (gasp, heresy!) mainframe enviroment. And before you go writing off that idea, consider that mainframes can handle 65535 separate IO devices. Refer to the article on IBM's channel architecture if you're interested. As everyone knows, a PC is lucky if it can handle 65 devices.

The PCI and Hypertransport approaches just won't cut it. Radical thinking (or reinventing, more likely) is needed with this kind of horsepower.

If Intel, or AMD, ever figures out how to handle that type of IO, instead of the rediculously small number of devices PC's are currently limited to, THEN we'll see a real shake up in this industry.

Re:Whoop-de do. More rethinking is needed. by rrohbeck · 2006-09-27 09:28 · Score: 1

With 80 friggin cores, you'd darn well better get the architecture right.

Well, they never got the Pentium 4 architecture quite right, and still sold boatloads of them. As long as marketing can sell more==better, we'll see a lot more data-starved cores.

IMHO, a moderate number of cores with specialized slave CPUs a la Cell would make a lot more sense for typical applications like games. 80 cores is pretty much MPP territory, so you need really specialized algorithms to keep them going, and 90% of those CPUs will be idle with your typical day-to-day computing or gaming.

--
thegodmovie.com - watch it
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-27 10:29 · Score: 1

By today's standards those interfaces were slow. Today's machines don't have need for 64K different I/O devices, but if they did they would have them. I/O interface addressability is the LEAST of the problem with the architecture of today's PCs.

How many hard drives can be attached to a single disk controller in a PC? A whole lot. Do you need more than that? If so, how many disk drives do you need? How many ethernet ports? What other high bandwidth I/O is there? None of this is the issue.
Re:Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-27 11:41 · Score: 0

"By today's standards those interfaces were slow."

IBM mainframes have the same modern interfaces like any other platform. The 64k devices that the grandparent quoted aren't necessarily physical devices. They're the abstraction that operating systems see. For example, you can have 336 4-gigabit fibre channel ports in a z9. After virtualizing these, you may use a big chunk of the 64k devices.
Re:Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-27 14:47 · Score: 1, Insightful

It's pretty clear you don't have the slightest clue what you're talking about. Nor has that stopped you from forming an opinion, and posting it too. :)

IO bandwidth is exactly the issue missing on PC class machines.

Do you have any idea of the setup time for DMA, and then interrupt overhead involved for each tranaction? For one disk, it's not a lot. Start driving "a lot" of disks so that you saturate the bus, it adds up. These are cycles that steal away from your CPUs main processing capability. Now add 80 CPUs, all starved for IO. It's a total waste of resources.

There's only so much that you can do in memory.

If Intel doesn't do something which can feed those CPU's constantly, you're going to end up with 79 space heaters, and 1 CPU. Given the crunch to conserve power, this clearly isn't feasible.

Let me put it simply for you. Without a major redesign, Intel is simply wasting its money. That's why IO bandwidth is a serious issue here. Without it, they are just talking out their butts.
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-27 16:48 · Score: 1

"IBM mainframes have the same modern interfaces like any other platform."

So there's no difference outside the size. Computers such as those are differentiated by RAS more than they are by I/O capability. PCs *could* be built with lots of bus and memory bandwidth.

"The 64k devices that the grandparent quoted aren't necessarily physical devices."

So that number is totally meaningless. It's nothing more than a 16 bit index.
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-27 16:53 · Score: 1

"It's pretty clear you don't have the slightest clue what you're talking about. Nor has that stopped you from forming an opinion, and posting it too. :)"

Plenty of that going around.

"Do you have any idea of the setup time for DMA, and then interrupt overhead involved for each tranaction? For one disk, it's not a lot. Start driving "a lot" of disks so that you saturate the bus, it adds up. These are cycles that steal away from your CPUs main processing capability. Now add 80 CPUs, all starved for IO. It's a total waste of resources."

Yeah, I do. What makes you think you know who you're talking to? This imaginary 80-core processor doesn't even exist yet, yet you claim to know that it is I/O bound. You are the idiot here.

"There's only so much that you can do in memory."

Yes, you can make it faster and add more of it.

"If Intel doesn't do something which can feed those CPU's constantly, you're going to end up with 79 space heaters, and 1 CPU. Given the crunch to conserve power, this clearly isn't feasible."

Perhaps you can lend your infinite expertise to Intel since you already know the achilles heal of their design. Processor cores that aren't doing anything aren't creating heat.

"Let me put it simply for you. Without a major redesign, Intel is simply wasting its money. That's why IO bandwidth is a serious issue here. Without it, they are just talking out their butts."

Thank you for talking so simply out your butt. It's a good thing we have real engineers in change of development.
Re:Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-27 19:02 · Score: 0

"What makes you think you know who you're talking to? This imaginary 80-core processor doesn't even exist yet, yet you claim to know that it is I/O bound. You are the idiot here."

I've got a pretty good idea of what kind of person you are, since you started off by stating something completely false. And then you followed up later by claiming the number of channel id's was totally meaningless (Hint: that's another strong statement you haven't the foggiest clue about this subject). And then there's your claim that making memory faster and adding more of it will solve everything. That's positively hilarious. Do you really think RAM is going to replace permanent storage anytime soon? Because unless it does, Intel is going to have to deal with I/O. This very simple fact seems to have completely escaped you.

If Intel had an IO solution here, they'd be crowing about that too. But the lack of their statements means their plans aren't far enough along; and the lack of it is glaring.

Just because you can type "make" doesn't mean you're either a kernel or a systems expert. 90%+ of the clowns calling themselves kernel hackers these days are just clueless. What you seem to have failed to realize is that you've made no technical statements to distinguish yourself positively in a technical light. So it's extremely clear what type of person I'm talking to.
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-27 23:34 · Score: 1

"...since you started off by stating something completely false..."

I thought that was you.

"...claiming the number of channel id's was totally meaningless..."

actually that was someone else. you have yet to prove that the current pc architecture is starved for device IDs.

"And then there's your claim that making memory faster and adding more of it will solve everything."

Didn't say that. I said the trend would be faster and more.

"Do you really think RAM is going to replace permanent storage anytime soon?"

Nope.

"Intel is going to have to deal with I/O. This very simple fact seems to have completely escaped you."

Yep, I fully realize that, but just made that up in order to argue with a ridiculous position.

"If Intel had an IO solution here, they'd be crowing about that too. But the lack of their statements means their plans aren't far enough along; and the lack of it is glaring."

No, it's not. If users wanted more disk I/O they'd put more disk in their machines. Intel doesn't control that, but Intel does sit on industry groups, such as SATA, that define future I/O interfaces. That group, BTW, is one that I personally worked on at one time.

"Just because you can type "make" doesn't mean you're either a kernel or a systems expert."

Same to you. I've written device drivers for multiple OSes, been release manager for multiple Unix products, been lead achitect for two RAID controller products, and been a senior member of technical staff for a Fortune 100 computer manufacturer (where I was the companies lead patent holder for several years). That WOULD make me a systems expert to some. What are your qualifications?

"What you seem to have failed to realize is that you've made no technical statements to distinguish yourself positively in a technical light."

Perhaps not to a moron like you, but at least I don't show myself to be a know-nothing.

"So it's extremely clear what type of person I'm talking to."

We're talking?
Re:Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-28 11:13 · Score: 0

"No, it's not. If users wanted more disk I/O they'd put more disk in their
machines."

Sigh. You are just absolutely full of silly clueless statements. One right
after another; you just don't seem to stop.

Adding more disks doesn't increase I/O. You can add 1 billion disks, and your IO
isn't going to increase. You'll hit a wall and be bus-bound. That's the
WHOLE POINT, which you just don't seem to comprehend.

80 CPU's are either going to be mostly idle, or be bus bound with the
current IO busses. In either former case, Intel's wasting money with their
current approach.

"That group, BTW, is one that I personally worked on at one time."

Ah, it figures. Most people on these groups excel more in paper pushing
and politics than they do technology. Typically they are second rate
technologists at best. This explains quite a bit.

Regarding your claimed experience, I'm completely unimpressed. I've done
far more than yourself, and on a wider scale. While some people would call
you a systems expert, you impress me as knowledgeable enough to be dangerous.

And yes, you are showing yourself to be a complete no-nothing, incapable
of basic comprehension of what you claim to have worked on.

Go study the mainframe IO architecture, and get back to me what you have
groked what real IO is all about.
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-28 14:44 · Score: 1

"Adding more disks doesn't increase I/O. You can add 1 billion disks, and your IO
isn't going to increase. You'll hit a wall and be bus-bound. That's the
WHOLE POINT, which you just don't seem to comprehend."

Talk about silly clueless statements. It's clear that I don't comprehend your WHOLE POINT because you don't have one.

Since you know so much about disk drives, please explain to me how many of today's disk drives it takes to saturate one of today's IO busses. You pick the drive and the bus. You won't be able to do it since you have no understanding of the subject. You're nothing but a poser.

"80 CPU's are either going to be mostly idle, or be bus bound with the
current IO busses. In either former case, Intel's wasting money with their
current approach."

First off, no matter how many CPUs you have and how busy they are, they will NEVER be IO bus bound because they don't sit on an IO bus. CPUs sit on a memory bus. With all your impressive experience, one would think you'd be able to get such simple terminology right.

Assuming you meant memory bus (which is a bog assumption considering your consistent demonstration of ignorance), you can't possibly know that because you have no idea what the memory bus throughput is. It's a safe assumption that such a CPU would be bus-bound using today's memory subsystems, but it's also a safe assumption that future systems will have much faster memory. It's an obvious requirement and one even Intel's janitor knows.

"blah blah blah"

Apparently you're main skill is to toss out meaningless, unsubstantiated insults as an AC. I noticed you haven't offered up any reason for us to believe anything you say. Maybe after another decade of education and experience you might aspire to be a second rate technologist yourself.

Why don't you tell us how "mainframe IO architecture" is going to solve this imaginary problem you know nothing about. Is it that Intel doesn't have enough device IDs? Perhaps if we only had 65535 channel IDs we'd all be in computing heaven.
Re:Whoop-de do. More rethinking is needed. by Anonymous Coward · 2006-09-29 03:49 · Score: 0

Sigh. You really have no idea of what you're talking about. IO bus saturation has always been a problem to contend with. This is why computer companies spend time and money dealing with this, in different ways. To proclaim that it isn't an issue is an even greater display of your cluelessness. Why do you think that companies design hardware in order to deal with this situation?

I can't believe anyone who has claimed to be on a standards committee can claim that it isn't an issue. You are clearly lying about your supposed background.

As for the number of disks, this appears in the Terabyte range. Especially for PCI and Intel systems, and even for AMD/Hypertransport systems, though it is higher there. Storage companies have had to explicitly address this issue. Take a look at what NetApp, Sun and Agami do to deal with it. While their approaches are on the cutting edge, this type of scale is now starting to come into play at the home PC market, with prices now down to under $1000 for a home version of a simple filer.

You're obviously lying about your technical capabilities, and are just a general moron who is trying to sound good. Unfortunately, it comes through loud and clear that you have no idea what you're talking about.

As I said before, every statement you make just digs you in deeper and deeper. Come back when you get a basic technical education.
Re:Whoop-de do. More rethinking is needed. by dfghjk · 2006-09-29 11:08 · Score: 1

If I could understand a single point you were trying to make then I could point out where you've gone hopelessly misguided. Unfortunately, all you are is bluff and bluster with a bunch of insults thrown in. This discussion has degenerated into a bunch of incomprehensable nonsense.

"As for the number of disks, this appears in the Terabyte range. Especially for PCI and Intel systems, and even for AMD/Hypertransport systems, though it is higher there."

WTF does any of that mean? Was that your sorry attempt to characterize disk performance as I requested? If it was, "Terabytes" aren't a measure of performance nor are they a measure of "the number of disks". Furthermore, PCI and Hypertransport are not equivalent technologies from Intel and AMD and the differences between the two have no impact on the number of disks needed in the system. Thanks for the effort, Curly.

"Storage companies have had to explicitly address this issue."

What issue?

"Take a look at what NetApp, Sun and Agami do to deal with it. While their approaches are on the cutting edge, this type of scale is now starting to come into play at the home PC market..."

What type of "scale"? What the hell are you talking about? Are you saying the home user now needs an enterprise NAS because his desktop processor is getting more cores? Are you crazy? You do realize that modest NAS boxes traditionally offer poor performance relative to direct attach storage, right? Do you think that anything ethernet attached will ever perform like SATA or SAS? What makes you think that the typical PC customer wants anything more than a single hard drive? Hell, mac owners are seemingly satisfied with a lowly 2.5" drive judging by the popularity of the (now dual core) mini.

Yes, we now see small NAS appliances for the home. They typically have one or two drives, are inexpensive, and are designed to be easily used. Performance isn't their strong suit. You want to throttle down your multicore PC of the future then force all of it's disk IO to run through an ethernet NAS. Good luck.

"IO bus saturation has always been a problem to contend with."

Of course it has been. That's why PC I/O busses have steadily improved over the years. Curiously, it is Intel that drives those improvements. Nevertheless, I/O performance is not effected by how many cores your CPU has. The original claim was that 80-core CPUs were a design mistake because somehow having that many cores will kill IO performance. That is, of course, bullshit.

"To proclaim that it isn't an issue is an even greater display of your cluelessness."

Good thing I didn't say that.

To paraphrase the thread, you contrasted an IBM mainframe architecture with the PC by saying that IBM could support 65535 devices where the PC would be lucky to support 65. I said device addressability meant nothing as PCs could hold as many IO devices as anyone wanted to put in them. You then changed the subject to IO bandwidth and I argued that CPU core count did nothing to discourage IO bandwidth. The rest has been groundless attacks on my qualifications (and all from an AC who can't even demonstrate that he understands the technology at all).

Desktop PCs have basic requirements. They typically have some kind of network connection whether it's ethernet or modem, they have some low speed peripherals like printers, card readers and HID, they have typically one hard drive and an optical drive, and they have a range of video capabilities from basic on up to workstation-class 3D. No amount of CPU development alone is going to change that formula and no psuedo-architect AC from the peanut gallery can formulate any argument for why a PC suddenly needs mainframe-class IO just because it's CPU gained more cores.

Someone needs to relearn SI by ncc05 · 2006-09-27 08:39 · Score: 5, Informative

[A] teraflop is approximately 1000 Megaflops.

Is there such a thing as a gigaflops? What happened to that?

Re:Someone needs to relearn SI by HatchedEggs · 2006-09-27 08:45 · Score: 1

Whoops, somebody slipped.

It's a common mistake... and it makes me feel special, because even my computer is faster than that.

Who needs R&D anyways?

Anyhow, we'll just send them the Wiki article on Teraflops: http://en.wikipedia.org/wiki/Teraflop

--
Justin - Don't be afraid of my blog, it won't bite.
Re:Someone needs to relearn SI by Ant+P. · 2006-09-27 11:05 · Score: 1

I think it's trying to say 1000 Itaniums run at a combined speed of one teraflop.

Redefining Tera? by backwardMechanic · 2006-09-27 08:41 · Score: 1

This must be some new definition of tera. From http://en.wikipedia.org/wiki/Tera

tera- (symbol: T) is a prefix in the SI system of units denoting 10^12, or 1 000 000 000 000.

That seems to leave Intel 999,999,999,920 cores short.

Re:Redefining Tera? by Anonymous Coward · 2006-09-27 09:00 · Score: 0

It's a teraflop computer, as in 1 000 000 000 000 floating point operations per second.

Erlang would shine! by Anonymous Coward · 2006-09-27 08:41 · Score: 1, Informative

This is the sort of hardware environment that would allow a programming language like Erlang to thrive. We could very well see it being used more and more often in the future, as even regular consumers have access to machines with well over 16 cores.

For many applications, Erlang provides a far superior model to that of languages like C and C++ (with pthreads), Java, C# and Perl when it comes to massively multithreaded programming. Very high reliability is possible using Erlang, as witnessed by the many telephony products in which it is used. So for consumer applications, it could help developers build very solid systems that easily take advantage of many processors.

Why stop there? by dtfinch · 2006-09-27 08:41 · Score: 2, Funny

Practicality and usefulness problems aside, you can fit over 6,000 6502 processors in the space of a P4, each running at several ghz.

Re:Why stop there? by Mixel · 2006-09-27 09:15 · Score: 1

So that's how Bender gets along with only 6502...

I have got to say by Philip+K+Dickhead · 2006-09-27 08:45 · Score: 1

That looked like it said Testicle computing.

--
"Speaking the Truth in times of universal deceit is a revolutionary act." -- George Orwell

Re:I have got to say by N+Monkey · 2006-09-27 20:39 · Score: 1

That looked like it said Testicle computing.
Well... I suppose that still works in binary... (except, allegedly, for a certain WW2 dictator)

Pentium - Stove-top model by diodeus · 2006-09-27 08:46 · Score: 2, Insightful

The first 80-core chip will actually look live a conventional kitchen hotplate. You add a pot of cold water on top of the chip, then with a dial on the unit you determie how much heat you want to produce. The CPU will automatically run the correct number of instances of Seti@Home to generate the desired level of heat.

The 4 X 80 "stove top" model will come out later that year. It will include an "oven" that has its own chip and convectional cooling.

Re:Pentium - Stove-top model by rco3 · 2006-09-27 08:56 · Score: 1

That's a funny comment, true. But it's not the worst idea in the world. People who use electric ranges spend hundreds of watts for nothing but heat - why not get some computation done in the process? SETI@Home, hack some Diebold machines, protein folding - whatever!

--

Ce n'est pas un vrai mouvement de robot!
Re:Pentium - Stove-top model by Anonymous Coward · 2006-09-27 12:14 · Score: 0

A water heater would be more useful - it's on more of the time and it's probably the most powerful heater in most households. Use it to decode HDTV and play games, with seti in the background to take the slack, and you have a winner (assuming you get the waterproofing right).

Still, I reckon testicle-roasting laptops are more likely to be the future... people do like them portables.
Re:Pentium - Stove-top model by daverabbitz · 2006-09-27 12:58 · Score: 1

Uhm, I've already been doing this for a few years. In the winter I power up my older machines and use them to heat and ventilate the basement.

Usually rendering, or just for show, but maybe if I can get some decent performance/watt I can get people to pay to heat my house :)

--
What could be better than a jet powered motorcycle? http://www.youtube.com/watch?v=u8l6GTHLSWE

More fifty cent words, eh? by robyannetta · 2006-09-27 08:46 · Score: 1

"Tom Yager over at InfoWorld imploring AMD to stop at 8 cores while everybody gets the architecture right."

I emplore Tom Yager to STFU and step out of the Intel/AMD competition. If Tom gets his way, neither chip vendor will see the newly paved CPU autobahn as an actual challenge. Each vendor has their own competing architecture and that's what makes things fun. If we were back in the older days of the dozens of i486 arch clones, things would not be so interesting. Anyone could stamp out one of those chips.

I think competition is a good thing. If AMD releases an 8-way, then Intel should release a 16-way. Let them compete. Consumers win with the low price wars they bring.

Shut up, Tom.

--
- Just my $0.02, take with a grain of salt, your mileage may vary.

Re:More fifty cent words, eh? by enrevanche · 2006-09-27 10:20 · Score: 3, Insightful

if you read his article, you would know that he wants them to create cpus that actually perform in the real world, not just add marketing numbers that will have very little effect.

Reminded me of Grade School by Shadow+Wrought · 2006-09-27 08:46 · Score: 2, Funny

You know the arguments in the yard at lunch:

AMD: We now have two cores, so there!
Intel: Oh yeah, well we now have four cores- losers!
AMD: Oh yeah, well we're coming out with eight cores next. Ha beat that!
Intel: We can and will! We're going to come out with, with EIGHTY cores! Yeah that's right, eighty cores!

Disclaimer: I've not kept up on the Core War, so any inaccuracies are for dramatic effect...

--
If brevity is the soul of wit, then how does one explain Twitter?

Re:Reminded me of Grade School by Anonymous Coward · 2006-09-27 11:07 · Score: 0

I've not kept up on the Core War, so any inaccuracies are for dramatic effect...

No, you pretty much got it right...
Re:Reminded me of Grade School by pipingguy · 2006-09-27 14:41 · Score: 1

Isn't there something called an Arm processor? Core vs. Arm - that brought back a lot of fond gaming memories.

Note an X86 but more of a super Cell. by LWATCDR · 2006-09-27 08:47 · Score: 1

The 80 cores are all simple floating point cores. A lot like the IBM/Sony Cell.
It is of interest for say super computers and video cards. It isn't the prototype of the Octodec80Core that will be in the new 72" iMac.
Yea it is a dupe alright.

What I think a lot of people are missing is that it almost looks like Intel is going to repeat the mistakes with Netburst all over again.
Now instead of a clock speed race Intel is starting a core race.
Intel is sticking more and more cores onto it's current FSB. This is going to bit them just like the clock speed race did.
Instead of the way to long pipeline of the P4 will have some really sick and twisted l1,l2,l3, and for all I know an l4 cache just to keep the cores from being memory staved.
If they do not watch it they will have 16 core systems that are slower than AMDs 8 core system. AMDs cpus scale better than Intel's thanks to the integrated memory controller and Hyper-transport links. I fear that Intel will have to follow AMD's lead yet again.

--
See my blog http://ilovecookes.blogspot.com/ for light hearted technical information.

Re:Note an X86 but more of a super Cell. by Doctor+Memory · 2006-09-27 09:32 · Score: 1

The 80 cores are all simple floating point cores. A lot like the IBM/Sony Cell.
Nope. That's what they made for the proof-of-concept demo, but ultimately they'll use something else. From TFA:
These cores will be low power and probably based on a past-generation Intel architecture that has been refined and perfected.
I'm betting Pentium Pro. Good, solid architecture, decent performance, still runs a lot of software. Wasn't the P-II basically just the PPro with MMX (and a half-speed cache...)? Hmmm, now that I think of it, parent could be right, they might go with the i860 or something like that.

Then again, since performance per watt is an issue, maybe they'll be able to go with a Pentium-M design. Not sure how power consumption differs from the PPro, and it's probably not comparable since I doubt they were ever both manufactured on the same process. Be interesting to see, from a technology standpoint. Anybody know how well Sun's high-end UltraSPARC systems are selling? Might help to indicate how much desire there is for big multi-core CPUs...

--
Just junk food for thought...
Re:Note an X86 but more of a super Cell. by LWATCDR · 2006-09-27 10:03 · Score: 1

Actually yes.
From the CNet story "Intel's prototype uses 80 floating-point cores, each running at 3.16GHz, "
So these are just floating point cores.
And from the TFA you your sited.
"the cores in a terascale processor will be much simpler (kind of like we are seeing in the Cell processor design)."
In fact it was the line above the one you cited.
Just like TFA said. A Super Cell.

--
See my blog http://ilovecookes.blogspot.com/ for light hearted technical information.
Re:Note an X86 but more of a super Cell. by dfghjk · 2006-09-27 10:34 · Score: 1

My guess is that Intel understands this seeing how they are already planning integrated memory controllers in future products. Funny how people here think they understand processor design better than the most dominant processor design company in the world. Memory bandwidth is something they've been dealing with for some time now.
Re:Note an X86 but more of a super Cell. by LWATCDR · 2006-09-27 12:56 · Score: 1

"Funny how people here think they understand processor design better than the most dominant processor design company in the world."
Then why did they stick with Netburst for so long?
I don't think you can call Intel the most dominant processor design company. IBM is killing them in the very high end with the Power 5 and beat them out with design wins in the XBox 360, Gamecube, and PS/3.
Intel doesn't produce the fastest CPUs or and it is very possible if you count all the PPCs even the most popular.
They just happen to have a good chunk of the PC marked.
Why do you believe that given their past history that they will not go for marketing flash over true performance? The Pentium D and the Quad Core are both clear cases of trying to have bigger numbers instead of true innovation.
The Core line is interesting and offers a real improvement but may also dead end like Netburst if they do not follow AMDs lead.

--
See my blog http://ilovecookes.blogspot.com/ for light hearted technical information.
Re:Note an X86 but more of a super Cell. by dfghjk · 2006-09-27 17:00 · Score: 1

If you want to argue against Intel being the most dominant processor manufacturer, go right ahead. You are welcome to look stupid but I won't play.

"Why do you believe that given their past history that they will not go for marketing flash over true performance?"

Intel's designs were always intended for true performance. Netburst was simply a failure in that respect. Intel believed at the time that it would scale.

"The Pentium D and the Quad Core are both clear cases of trying to have bigger numbers instead of true innovation."

Bullshit. What is "true innovation"?

"The Core line is interesting and offers a real improvement but may also dead end like Netburst if they do not follow AMDs lead."

Yeah. You can bet Intel is following AMD's lead.
Re:Note an X86 but more of a super Cell. by Anonymous Coward · 2006-09-27 19:17 · Score: 0

"Intel is sticking more and more cores onto it's current FSB."

Yet another person who have not followed what Intel has said.
Intel will begin to introduce integrated memory controllers and routers in 2008.

Re:They are preparing by Sillygates · 2006-09-27 08:49 · Score: 1

Now, imagine a beowulf cluster of those NOT running one instance of Vista!

--
I fear the Y2038 bug

Lots of uses for 80 processors by gsfprez · 2006-09-27 08:50 · Score: 4, Interesting

I'm a video guy. I can't render video fast enough. I can't do transcoding fast enough. My video is getting larger and deeper in color, and i need more power.

all of that is threadable.

so is photographic processing. You can divide a picture 80 ways and have each processor do whatever it is you want to do on it.

Gamers? Fscking a.... i'm so SICK of hearing hiow everything is for them. Just because something isn't going to help Halo Life 3 run faster is not any of my concern.

There are lots of people working on their computers that want to see more cores because it will make our lives better.

--
guns kill people like spoons make Rosie O'Donnell fat.

Re:Lots of uses for 80 processors by MrSteveSD · 2006-09-27 09:49 · Score: 1

I agree. You can never have enough processors. I do 3D rendering and that is massively parallelizable. Also once the power is there, new things will emerge that use it in a constructive way. For example, when computers became powerful enough, suddenly people were using them for video editing, and that really does require the power.
Re:Lots of uses for 80 processors by HerbieStone · 2006-09-27 23:41 · Score: 1

Ah but you look at it the wrong way.

I've seen the progress on stuff that weren't for gamers. FireGL-cards for video-processing and their ilk. This stuff gave you more power but were expensive... and stayed expensive for years to come without much progress. The Gaming-rigs are used by a wide audience, there is a big market and have multiple competitors, which over time results in a lower prices and faster new hardware.

So be happy when your hardware can also be used by gamers, it will soon cost a lot less and/or something better can be expected in a year or less.

Greets
HS

Heeeere we go again. by tygerstripes · 2006-09-27 08:50 · Score: 4, Insightful

I'm sorry but, well... didn't you guys do this with processor speed a while ago?

That didn't work because AMD worked out that architecture can trump speed. They innovated, and then did it again with decent dual-core (as in NOT the two-dies-on-one-chip cack that you churned out at first).

So, you improved your architecture and implemented dual-core properly, to produce the fantastic Duo. You got back in the race.

And then there was talk of more cores. And you went "Fuck that, bitches, stay DOWN - we is gon' fuck you up good with 80 cores, bitch, an' dat hard!". Yes, you decided to try and dominate the pissing contest of multi-core instead of megahurtz.

Jesus guys, didn't you learn a fucking thing? STOP trying to turn out something that little bit "more" than the competition, just get on with innovating and coming up with damn good chips. That's how AMD threatened you and, if you go on with this "anything you can do" shit again, you'll be back to square one.

--
Meta will eat itself

Re:Heeeere we go again. by Psiren · 2006-09-27 09:37 · Score: 1

I'm taking the 80 cores thing with a huge pinch of salt, as I'm sure everyone else is. Many have pointed to the whole "Pentium 4 will scale to 10GHz" debacle as an example of Intel making dubious claims.

However, I've not read anything that would suggest Intel isn't looking at improving memory and bus throughput. They'd be mad to think they can stick 80 cores into an existing system and have it do anything useful. I don't think even Intel are stupid enough to try and get away with that one.

I don't think 80 cores is feasible in the near future, but as long as they keep pushing the envelope and forcing AMD to compete technically, that's just fine by me.
Re:Heeeere we go again. by mudetroit · 2006-09-27 10:04 · Score: 1

Oh good lord, at least take the time to really find out about something before insulting something. The 80 cores they are talking about are simple floating point cores, not full featured ones like in the Core 2 or AMD64. They were actually throwing out the idea, in as much as anything, as a compliment to IBM/Sony for the ideas the the Cell processor brought forth. The thing that it points more towards is that they are considering different ways of approaching problems then simply tacking more highly complex processors onto the system, which after a point has very questionable merits.
Re:Heeeere we go again. by tygerstripes · 2006-09-27 20:04 · Score: 1

Hmm, fair point - I didn't look into it much. I do sincerely hope this is a genuine innovation though, and not just an attempt to wow the media with the latest buzzword. Let's be honest, the possibility of being in the spotlight can't have been far from Intel's mind when they announced this...

I was genuinely excited when the Cell processor was being talked about, but they seem to have screwed up the commercial implementation somewhat. I was just as excited about the Crusoe architecture all of 5 years ago, but that never even made it out the laboratory door.
If Intel really do take a great new idea and make it work, hats off to them. I'm just wary of the ol' dinosaur falling back into old habits, as it's very easy for big-big companies to do.

--
Meta will eat itself
Re:Heeeere we go again. by a.bokovoy · 2006-09-27 22:29 · Score: 1

So, this would finally be what 86 in x86 mean, 86 cores.

FSB bandwidth first by Anonymous Coward · 2006-09-27 08:52 · Score: 0

I would prefer to see more bandwidth on the bus then more cores/gigahertz in the CPU. Let's get away from 15x multipliers and get back to 2x multipliers, bring the rest of the system more in line with the CPU before we start scaling the cores out.

Obligatory Duke Nukem Forever Comment by Pharmboy · 2006-09-27 08:53 · Score: 1

oh oh oh, so THAT is what they are waiting for before releasing Duke Nukem Forever? Eye candy that requires a minimum of 80 cores MUST be good!

--
Tequila: It's not just for breakfast anymore!

Re:Obligatory Duke Nukem Forever Comment by Anonymous Coward · 2006-09-27 09:28 · Score: 0

When its done LOL I sure Hope I live that long. I wonder if my Social Security check will cover the cost of the game

Good information, but little value to the... by HatchedEggs · 2006-09-27 08:55 · Score: 1

average consumer. Lets face it, most of us aren't going to see anything along these lines for quite some time. So while I give a nod at the excitement surrounding the technical research going on here, it isn't going to do much for me having to run bloated software on my work computer. And we certainly aren't going to see a "Beowulf full of these" anytime soon.

That said, I think that the benefit these will have on the scientific community (as well as servers, etc potentially) will be quite high and some time in a dozen or so years there will be practical products developed from this technology.

For some reason when I saw this initially I wondered if they might have some kind of potential use in artificial intelligence. For one the processing power available, but perhaps the ability to leverage the various cores will bring about some efficiencies that make it more practical.

--
Justin - Don't be afraid of my blog, it won't bite.

Memory busses are for swapping by Sloppy · 2006-09-27 08:57 · Score: 4, Insightful

Tom Yager writes:

If I had a vote, I'd have both vendors stop at four cores and focus on fat and fast busses that give those cores something to fill instead of something to wait for

What's a memory bus? Oh right, that thing you use to access the DDR4 swap device when the page you want to access is no longer in the on-CPU RAM. ;-)

Seriously, look at the growth of L2 caches, and tell me the day isn't coming when they just call it "RAM" instead of "cache." If Intel and AMD want to keep piling transistors onto their chips, this'll give 'em something to do.

--
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.

I don't get it... by Anonymous Coward · 2006-09-27 09:02 · Score: 0

...why is everyone so nut's about the CPU it's just one part of the whole thing. A central part granted, but CPU capacity currently isn't the bottleneck or did I miss something.

I am not good at the hardware stuff but multiplying the cores again and again does not seem to be a revolutionary strategy. Or is it? (That's a real question.)

Or do we get all this core race just because that's were the publicity is and AMD and Intel need the media attention for their stock-market price?

640 cores by ion_ · 2006-09-27 09:06 · Score: 5, Funny

640 cores should be enough for anyone.

Re:640 cores by Quaoar · 2006-09-27 09:28 · Score: 1

I think I've heard this joke 640 times, and that's definitely enough for me...

--
I'll form my OWN solar system! With blackjack! And hookers!

Zen exercise... by Anonymous Coward · 2006-09-27 09:09 · Score: 0

Zen exercise for today:

"Wow, don't imagine a Beowulf cluster of these!"

Arrgghhh by Usquebaugh · 2006-09-27 09:12 · Score: 3, Interesting

Why is it that with intel talking about a radical change in consumer hardware the level of comments on /. is barely higher then that on AOL.

We have had multi processor machines for ages. This is not a sudden unknown. Look up transputer, connection machine, beowulf, cray. There is still ground to be covered but it's not unkown territory. The difference is this is intel, intel needs a big market to sell to.

This is not going to make significant difference to the end user, most of them will still write letters, calculate spreadsheets and browse the web. It might be enough to finally expose MS et al for what they have always been, the parasites.

Where this is going to hit home is in the realm of programming and OS.

Want to run an OS primarily designed for uniprocessing on a multi way architecture? Look at the issues Win&Lin have with SMP, limited to 16 processors I believe. Numa and beowulf are a different kettle of fish. So what will we have on these massive SMP architectures?

Programming, at last we might be getting out from under VonNuman. Progress might be possible after 30+ years of stagnation. The symbolic/functional languages are going to start to move forward. Hell we might even get to run on stack based cpus with energy reclamation automated :-) Of course a nice message passing symbolic language might score big.

But given then history of software we'll have a bunch of ignorant, loud mouth idiots running around telling everybody the one true way is Java with mutex and semaphores. PHBs will grab at the first thing that has enterpise written on it and is 'guaranteed'. Most programmers will code how they have always coded head down, ass up. The number of processors will double every two years and the speed of software will continue to halve in the same period.

Of course nobody will suggest that a staged conversion should take place. There will be all these reasons to throw everything away and start over. Because this time we'll get it right!

Re:Arrgghhh by dfghjk · 2006-09-27 10:44 · Score: 1

"Want to run an OS primarily designed for uniprocessing on a multi way architecture? Look at the issues Win&Lin have with SMP, limited to 16 processors I believe."

Neither Windows (NT+) nor Linux were "primarily designed for uniprocessing". Both platforms run on much larger than 16-way systems. Licensing issues are a different matter.

"So what will we have on these massive SMP architectures?"

Applications that won't use them for a while?

"Programming, at last we might be getting out from under VonNuman."

What does that mean? Are you saying you've "been under" "VonNuman"? I hope he's hot.

"Hell we might even get to run on stack based cpus with energy reclamation automated :-) Of course a nice message passing symbolic language might score big."

If history teaches us anything, we should expect to continue running x86.

"But given then history of software we'll have a bunch of ignorant, loud mouth idiots running around telling everybody the one true way is Java with mutex and semaphores. PHBs will grab at the first thing that has enterpise written on it and is 'guaranteed'. Most programmers will code how they have always coded head down, ass up. The number of processors will double every two years and the speed of software will continue to halve in the same period."

Some of us have been around long enough to consider that new-fangled. Java is a lot newer that Unix is yet Linux is our saviour.

"Of course nobody will suggest that a staged conversion should take place. There will be all these reasons to throw everything away and start over. Because this time we'll get it right!"

Enjoy going bankrupt while getting it right. Those with the money set the rules.
Re:Arrgghhh by Procyon101 · 2006-09-27 11:08 · Score: 1

Yes, but in the background, CS will continue to move foward, in fits and false starts as it always has of course, but forward, by the real researchers working in the background and totally ignored by the mainstream. The maths will be better and better understood over time and significant proof are completed. Once in a while one of the enterprisey languages will pick up one of their ideas, call it the holy grail, and introduce it to the ass up programmers forcefully, which they will find uncomfortable for a while, but will eventually grok and get used to. In the mean time, the Mutex based C# VM that needs to copy significant portions of itself into core specific RAM to be able to handle such concurrency, causing artificial inflation of requirements for harware will fund the continuing exponentially growing hardware pissing contest with capital it wouldn't otherwise have if the pointy-haired bosses ever realized they could do what they're attempting on a palm pilot if they understood computing logic in the slightest. This pissing contest gives resources to the the CS people by commoditizing hardware that would otherwise cost them millions. Occassional, a couple CS guys looking to get laid will their "new technology algorithm", lambda-foldr or somesuch other 1st year CS concept that's actually designed to run in parallel, and will get rich.

Look at the bright-side. The more wasteful of proc cycles "industry" is, the more horsepower we get subsidized. The fact that industry is being dumb has nothing to due with acedemic stagnation... acedemia stagnates because it occassionally crashes headlong into hard problems and needs a bonifide genius to break through with a new, revolutionary idea, along with a gob of low hanging fruit to be grabbed along with it. Industry isn't going to come up with this breakthrough, IMO, it's too abstract and seemingly impossible. The breakthrough will come from a single zit-faced kid on his laptop with too much time on his hands to think with, we will all call him a moron for his stupid ideas, and then we will shut up to a deafening silence when we realize what he did; just like always. The market will catch on to the idea 20 years later, in an impure form that management can stomach.
Re:Arrgghhh by Usquebaugh · 2006-09-27 11:21 · Score: 1

Which large scale SMP system does Linux & Win run on? I thought the HAL for Windows was designed for 16 way and that Linux SMP is being added as needed. The issues of SMP are prevalent on both systems, when devloping the OS the multi processing was not high on the priorities.

Von Nueman(sp) is the uniprocessor architecture. Single bottle neck through which all data and instructions must pass.

I remember a time before x86 and the current x86 chips are just x86 on the surface. The x86 instruction set is split up at a lower level to allow pipe lining etc.

I'm not sure tenure counts for anything in computing but if it does you're pretty dumb claiming it anonymously, I've now got my fingers crossed your not some big name CompuSci guy :-)

Given that most of the CompuSci papers that deal with the big picture had been written by 1960, any body could read the fundamental papers in under a year and have a damn good understanding about computers. Now I'm not saying that all valuable research had been completed by 1960 but the vast majority had been. I would say OO, Packet Networks & non Von Nueman machines are areas that came after 1960 and could be classified as ground breaking. I'm hard pressed to name anything else?

Linux is nobodies saviour. It's an OS. It's my OS of choice and has been since '96 but that doesn't mean it's anything special. I might switch to another GPL OS in the future, Minix seems the best bet, but maybe Plan9 or some OS designed for multiprocessors. I doubt I'll be running Linux in 5 years!

Those with money set the laws and only fools are ruled by any of those three.
Re:Arrgghhh by Usquebaugh · 2006-09-27 11:40 · Score: 1

Although I tend to agree with most of the above. I am concerned that nowhere is CompuSci blue sky research being done. Where is Princeton of the 40s, Burroughs of the 50s, Bell Labs of the 60s, Xerox Parc of the 70s? Yes I did sort of make up the dates etc to make a point but it's still pretty close.

Today it seems that all research must show a profit/product. Is any where looking to hire the cream of post docs, those with radical ideas and the skills to implement. How about tackling a big problem e.g. the first generally available quantum computer, the first self aware AI? Instead we have supposedly intelligent thinkers working on a cheap laptop, ohh look it's lime green, radical man! AI research more interested in mechanics than biology.

I would love for a goverment or a company to challenge a hard problem e.g. landing a man on the moon, we had a pretty good idea how to get there we just needed the impetus to go. Now Bush would not be ideal figure head for AI research. But Google, setting aside a billion dollars and 100 postions to tackle quantum computing would be a landmark idea, fixed term of five years now off you go.

Could we the great unwashed fund such a project through a non-profit? Linux users for AI or some such. Rather than the big splash of google the slowly building wave?
Re:Arrgghhh by Procyon101 · 2006-09-27 11:56 · Score: 1

We knew how to get to the moon roughly. We know how to buld Quantum computers roughly. We haven't the foggiest idea of how to make a strong AI. No amount of money in the world will make that strok of genius happen... it just will.

To answer your questions though, the reasearch is being done in the universities mostly, supplimented by the big corps. Microsoft has some excellent reasearch going on in this field that I have seen first hand, as does IBM and Google to name a few. The missing peice is still that Einstein/Turing like figure that will come up with the revolutionary spark of an idea that will get the ball rolling. Without that seed, we are just refining the old ideas.
Re:Arrgghhh by Usquebaugh · 2006-09-27 12:52 · Score: 1

Which university, which papers? I have yet to see anything with MS on it that makes me do anything other than yawn. IBM got mired in process refinement during the 90s, make the same stuff better. Google I'm not to sure about, they have a lot of people with credentials but they seem to go strangely silent when they work for google?

Nothing wrong with academic research but it's rarely blue sky and if it's being funded by a business it's usually difficult to get info. Prehaps I'm looking in the wrong arenas but ACMS/IEEE plus top ten CS Universities plus the research arms of IBM/MS etc never seem to show anything radical. If I have to read one more paper on OO-RDBMS mapping or network agents I think I'm going to scream.

Now I'm far from being and academic and I'm not qualified to judge the work in any meaningful way other than with my own prejudicies. So prehaps I'm being too hard but the more phd dissertations I read the more convinced I am that the papers are being deliberately written in a confusing manner with the idea being if they throw in enough long words then they'll get the phd from a submission hold or knock out rather than an original idea that significantly extends knowledge.

Prehaps, I'm just becoming old. But reading Church, Turing, Shannon, McCarthy et al Is enjoyable, prehaps because I know the value of the work, prehaps because I can follow most of it, I don't know but I read most modern work and feel as if I'm being insulted. Almost a case of 'yeah so what, show me the breakthrough, show me the new idea, new application'.
Re:Arrgghhh by Procyon101 · 2006-09-27 13:11 · Score: 2, Insightful

The complexity of the research is due to the fact that it is so very, very refined. There isn't much out there that's revolutionary, so instead of "Look at this great concept... a wheel" we are getting "The benefits of XYZ fractal tread on a vulcanized rubber tire under wide tempurature variations spanning water phase changes." The second paper is much more educated on the subject, but much drier and not at all revolutionary ;)

In Genetic programming, we had heirarchial GP a couple years ago, breaking through the long-running GA problem we have suffered from for 20 years which renders GA useless for anything but trivial search spaces. This was a brain-dead, forehead slapping stupid solution that had just been missed by everyone. MIT press broke the paper, the widespread implications are not yet being felt. This is just a simple example, but hopefully this unlocks some bigger solutions.

Type-calculus is still a very hard unsolved problem, even though the concepts of data types have been around longer than computers. A refined calculus here will open up alot of doors. There are alot of people working on it.

But yeah, we haven't had a Turing come through in quite a while, and it's about time. I just don't think you can "buy" genius. Throwing a bunch of money at the problem may just be throwing it to the wind, or it may indeed foster an environment that this generation's Church comes from... there is no way to know.

Oh, and I can't comment on MS's research... they'd be mad :) But there is some Blue Sky research going on.
Re:Arrgghhh by David+Greene · 2006-09-27 14:32 · Score: 1

Which large scale SMP system does Linux & Win run on?

Your question doesn't make sense. An SMP architecture itself is very limited in the number of processors it can efficiently support. If you're asking which large scale systems Linux runs on, the answer is "lots." Most of the big supercomputer manufacturers are migrating to Linux, for example.

I would say OO, Packet Networks & non Von Nueman machines are areas that came after 1960 and could be classified as ground breaking. I'm hard pressed to name anything else?

Caches, virtual memory, multithreading, SMT, prediction (all kinds), multiprocessing, integrated circuits, LSI, VLSI, CMOS, optical interconnect, ...

"Ground-breaking" change is very rare. I would argue that OOO was not all that groundbreaking. It's a natural evolution of sequential computing. I'd put it in the same class as pipelining. All of the things I mention above are at least as important.

--
Re:Arrgghhh by David+Greene · 2006-09-27 14:43 · Score: 1

Prehaps, I'm just becoming old. But reading Church, Turing, Shannon, McCarthy et al Is enjoyable, prehaps because I know the value of the work, prehaps because I can follow most of it, I don't know but I read most modern work and feel as if I'm being insulted. Almost a case of 'yeah so what, show me the breakthrough, show me the new idea, new application'.

How long have you been reading papers? A wise man once remarked that grad students go through three phases: the "OMG everything is great!" phase, the "Geez, all these papers are stupid" phase and finally reach the, "Ah, I can recognize what's good and what's crap" phase.

Most research papers are about evolutions of existing ideas. That's how research works. You will very rarely read a new paper and think it's a revolution. I haven't yet come across one. You cite the big names because of the big papers they wrote, but how many "ordinary" papers did they write before and after the big one? Lots.

There's a huge step from the research lab to production computing. Functional languages, for example, are used in lots of specialized applications. But they aren't suitable for general-purpose computing because no one has figured out a good way to run them efficiently and write the software we have without destabilizing the entire world of computing. That's life.

Many paces are doing good computing research. You've just set your expectations much too high.

--
Re:Arrgghhh by Usquebaugh · 2006-09-27 15:15 · Score: 1

Cache/Virtual memory were hardly revolutionary, think of a turing machine and it's almost impossible not to see that memory could be virtual.

IC/LSI/VLSI/CMOS are vital to modern computers but not to the theory of computers. Might as well have steam powered cpus.

Optical interconnect, now your grasping.
Re:Arrgghhh by ekhben · 2006-09-27 15:56 · Score: 1

There will be all these reasons to throw everything away and start over. Because this time we'll get it right!
You said this, and didn't get moderated Funny? Well done, sir, well done.
Re:Arrgghhh by bfree · 2006-09-27 16:26 · Score: 1

Linux SMP is being added as needed
Yep, as needed ... do you need more then 512 processors under one instance of Linux and as much as 128TB of globally shared memory yet?

--
Never underestimate the dark side of the Source
Re:Arrgghhh by dfghjk · 2006-09-27 17:06 · Score: 1

"Which large scale SMP system does Linux & Win run on?"

What? Are you kidding?

"...the current x86 chips are just x86 on the surface."

Bullshit. x86 is defined by its instruction set. In that respect, the G5 is only "PowerPC" on the surface, too.

"I'm not sure tenure counts for anything in computing..."

Then you are a fool.

"...but if it does you're pretty dumb claiming it anonymously..."

what?

"...I've now got my fingers crossed your not some big name CompuSci guy :-)"

what? You'd hate to argue with one of those?

"Those with money set the laws and only fools are ruled by any of those three."

Apparently you're still in school. Hopefully you'll understand computers when you grow up.
Re:Arrgghhh by Usquebaugh · 2006-09-27 17:56 · Score: 1

NUMA not SMP
Re:Arrgghhh by Usquebaugh · 2006-09-27 18:10 · Score: 1

No I'm not kidding, give me one example of a Linux or Windows running on a 16 way SMP box.

It's been talked about how do we identify cpus. You would go with the external interface. Anything that accepts x86 code is a x86 cpu. Intel, AMD...Transmeta etc. If an external instruction set is deconstructed to lower level op codes is the cpu still an x86. Take a look at the P4 design docs, they talk about micro op codes as do most cpus that try to maintain backward compatability for their instruction set.

I'd love to learn/dicuss from/with some of the big names pretty sure an argument is not what I ever want.

Would seem that being unable to attack the point you attack the person :-) Been working 20+ years in computing as a developer. I'll grow older but not up, if you ever understand why you might stop being so rude when you have a perceived cloak of anonimity to protect you.
Re:Arrgghhh by Anonymous Coward · 2006-09-27 18:39 · Score: 0

Thanks for the pointer to hierarchical GA. I have some ideas about taking advantage of hierarchy in local search and have been wondering why the principle of hierarchy is so little used in literature. At least the idea can't be half bad if there's a clique of researchers working on it... There are so many angles on attacking combinatorial optimization problems that it's impossible for one person to be aware of the state of the art on all of them.
Re:Arrgghhh by Procyon101 · 2006-09-27 18:44 · Score: 1

FYI, if you want to stay on top of the state of the art, MIT Press has a "Evolutionary Computation" Journal bi-monthly which has the current state of the art papers, and is only around $70 a year subscription.
Re:Arrgghhh by Anonymous Coward · 2006-09-27 21:27 · Score: 0

Why be so dismissive of NUMA? So it takes a few more cycles to access non-local memory. That's just a fact of life in a lightspeed-bound universe, and from the point of view of a programmer, all memory is the same speed within a small constant factor. On the other hand, if constant factors matter, then you should be ecstatic about the impressive improvements in chip manufacturing that have increased software speeds by several orders of magnitude. Care to elaborate on your point of view?
Re:Arrgghhh by dfghjk · 2006-09-28 00:04 · Score: 1

"No I'm not kidding, give me one example of a Linux or Windows running on a 16 way SMP box."

I don't need to. An example was already posted by another before you even posted this question. http://www.sgi.com/products/servers/altix/4000/

Why did you ignore this example?

Here's the Windows 2003 spec. Up to 64-way: http://www.microsoft.com/windowsserver2003/evaluat ion/sysreqs/default.mspx

Windows NT was SMP from the very beginning.

"It's been talked about how do we identify cpus. You would go with the external interface."

Of course I would. So would any other sane person. The instruction set defines what the processor does.

"If an external instruction set is deconstructed to lower level op codes is the cpu still an x86."

Yes it is, provided it's the CPU doing the deconstruction. Instruction decode is the first step of execution in any processor. Doing so doesn't somehow make the processor "not x86".

"Take a look at the P4 design docs, they talk about micro op codes as do most cpus that try to maintain backward compatability for their instruction set."

Yeah, so? Are you arguing that NetBurst is really a different processor masquerading as x86? It isn't. It was specifically designed to run x86 and nothing else.

Let's look at this another way. If a processor runs x86 code without emulation then it's x86. There are two processors that don't directly run x86 that are worth considering. Transmeta is the first (and they made more than one). Transmeta had emulation software that was loaded to run x86 and could theoretically run other instruction sets or even it's own internal one. Thing is that never happened. The other is Itanium. Itanium ran x86 only partially in hardware and required software emulation assistance. All other x86 processors execute x86 code natively starting with the first instruction coming out of reset. If that isn't x86 then I can't imagine what would satisfy you.

"...you might stop being so rude when you have a perceived cloak of anonimity to protect you."

I have no more a cloak than you do. I'm not posting as an AC and I'm not the only one hurling insults. Let's just say "I've now got my fingers crossed your not some big name CompuSci guy :-)" Smiley indeed.
Re:Arrgghhh by Anonymous Coward · 2006-09-28 02:58 · Score: 0

Thanks again, I'll check it out... In the few GA papers I have seen, the thing that struck me was the modelling approach, which seems to be based on conditions that allow the GA to work, and there is comparatively little focus on how well actual problems fulfill those conditions. To strengthen my prejudices, the first hierarchical GA paper I found started by presuming that there are epsilon-optimal subproblem solutions that can be constructed into a global solution, and working up from there.

What I'd like to see is a more rigorous approach, starting from a standard NP-hard problem and demonstrating that the hierarchical algorithm provides a solution within certain bounds of the optimum. Would you happen to know of any work along those lines?
Re:Arrgghhh by Usquebaugh · 2006-09-28 04:19 · Score: 1

I did not ignore the example it's a numa not a smp box!

Like I said the x86 doesn't run x86 under the covers it just presents an interface.
Re:Arrgghhh by dfghjk · 2006-09-28 06:12 · Score: 1

You may have said it but it's wrong. x86 processors are defined by the code they run just like all other processors are. No x86 processor EVER executed every x86 instruction directly and that includes the original 8088/8086. All modern processors work internally like x86 processors today, but the internal design of the execution engines have to be tailored to the instruction sets that are being executed.
Re:Arrgghhh by Procyon101 · 2006-09-28 09:31 · Score: 2, Informative

To clarify, by heirarchial I was refering to HFC.

GA/GP searches a problem space for answers using a distributed hill climbing algorithm. It isn't a magic bullet for all problems, but performs well when the fitness of the solution set is a contiguous function and the slopes of the fitness hyperplane are not too extreme. If the fitness landscape is not contiguous, then GA/GP is unlikely to outperform random search by very much.

For instance: If your problem is "Devise a key that will open this lock", then GA/GP is not a very good way to go about searching because the feedback for whether the solution is good or not is purely binary... it either works, or it doesn't. In a failed attempt there is no "hints" as to whether this failure was closer of further away from any solution. If, however, we can gague how good the solution is in comparison with other attempts, we can hill-climb the fitness landscape and home in on good solutions.

Traditionally, however, GA/GP has been limited by the fact that it tends to home in on 1 good solution, to the exclusion of all others, even if it's not the "best" solution. The algorithms tend to refine a good solution forever, never escaping their local maxima in the fitness landscape. The only way to get them to find another solution is to restart the evolution from the beginning. HFC allows the run to continuously probe the entire solution set and converge on all maxima.

Your question about optimal subsolutions is debated actually. Koza's original premise was that GP works by combining good subtrees in a solution. More recent research has brought this into question, and often random mutation will outperform crossover, or at least come close. There is work being done to see if we can increase the role of solution "building blocks" but there is no concensus. Your NP problem does not need to be highly decomposeable for most GA/GP systems to work, but it does need to have a smooth fitness landscape.
Re:Arrgghhh by Procyon101 · 2006-09-28 10:42 · Score: 1

I've run Windows and SQL Server on a 64 proc intel box. It was a monster machine used internally for testing.

It kicked the monkey's buttocks too, btw.

Is it warm in here or is it just me? by psbrogna · 2006-09-27 09:14 · Score: 1

I expect we're going to need some progress from the fan guys.

Bah. Typo. by Anonymous Coward · 2006-09-27 09:15 · Score: 0

"Refer to the article on IBM's channel architecture if you're interested."

I had meant to refer to the article on wikipedia. They have a good overview of the approach. Supposedly, IBM is now moving completely away from interrupts as well, as they take too much overhead with that number of devices; and are instead moving towards a polled approach. At least that's what I've heard; I haven't seen this for myself, so I can't confirm it.

These are interesting times we live in.

Reminds me in a way of the EFF's Deep Crack by HatchedEggs · 2006-09-27 09:18 · Score: 1

Having all the cores on one board like this rather reminds me of EFF's Deep Crack.

http://en.wikipedia.org/wiki/EFF_DES_cracker

I know it is a completely different thing, but the EFF put a bunch of custom chips on boards (60+) and could crack the "then" nearly uncrackable DES crypto in a matter of 5 days.

That said, anybody want to bet the NSA is going to be the first people on the list when these things become practical in another couple years?

--
Justin - Don't be afraid of my blog, it won't bite.

Re:Reminds me in a way of the EFF's Deep Crack by snp-7-3 · 2006-09-27 12:06 · Score: 1

The NSA already has them... where's my tin-foil hat...

400 might be better by Anonymous Coward · 2006-09-27 09:28 · Score: 0

Depending on the nature of your workload, a 400 object Field Programmable Object Array (object ~ core) might be a better choice than a tradition CPU.

Now we can have 80 DUPs at the same time by EmbeddedJanitor · 2006-09-27 09:30 · Score: 1

Didn't we all argue about this already yesterday.

--
Engineering is the art of compromise.

Re:Now we can have 80 DUPs at the same time by dc29A · 2006-09-27 12:16 · Score: 1

Didn't we all argue about this already yesterday.

Thank $Deity$ for dupes. I forgot to cancel my pre-order for a 10 ghz P4 CPU. But thanks to this dupe, I did cancel it and instead placed a pre-order on an 80 core CPU. YAY!

The race to the platter. by Anonymous Coward · 2006-09-27 09:37 · Score: 0

Bandwitdh, shmandwidth.
If data I/O is bound to a rotating disk platter, there's yer bottleneck.

Where's my 200 + GB RAM Array that loads the entire hard drive into memory, runs everything from there, then writes it all back to the hard drive whenever it has a moment (or just before shutdown)?
CPU, cores, bus bandwidth, bus speed, reduced memory latency and all that is great, I mean really great. But if it all eventually has to wait on a mechanical hard drive, who cares?

I can see it all now . . . "My system can wait on its hard drive I/O faster than your system!"

OK, I'll bite by smcdow · 2006-09-27 09:46 · Score: 1

... unleash a new era in data-mining, ...

Data mining? How does having 80 cores improve I/O?

--
In the course of every project, it will become necessary to shoot the scientists and begin production.

Re:OK, I'll bite by Anonymous Coward · 2006-09-27 10:15 · Score: 0

Why is improved I/O necessary for a new era in data-mining?

Min Req by Wiarumas · 2006-09-27 09:54 · Score: 1

Perhaps someday a game will have the minimum requirements of an 80 core processor.

--
I will bend like a reed in the wind.

Haven't you heard? by TimothyTimothyTimoth · 2006-09-27 09:56 · Score: 1

Gillette is approaching a Singularity... http://www.collisiondetection.net/mt/archives/2006 /06/the_gillette_si.html

--
It doesn't matter which ape activates the Monolith

So does this mean... by Anonymous Coward · 2006-09-27 10:02 · Score: 0

...that the average (l)users comp that I have to fix wont be frozen from the 76 processes of spyware running and that I can actually use the system and repair it there?

"For balance" by Overly+Critical+Guy · 2006-09-27 10:05 · Score: 1

Why did kdawson add that to the submission? Since when has Slashdot done that for AMD press releases?

--
"Sufferin' succotash."

Doesnt matter by Lonewolf666 · 2006-09-27 10:10 · Score: 1

The average buyer will not understand "out of order architecture" anyway. The MHz race was different, because even non-techies could see how the computers got faster with increasing clock speed.
But now?
Maybe it will be "number of cores". Otherwise Intel and AMD will have to use meaningless slogans like "Intel inside" to suggest a sense of security when using their particular brand.
I expect a mixture of touting lots of cores and almost-fraudulent crap like "the Pentium III will make your internet faster".

--
C - the footgun of programming languages

Memory bandwidth by rbanffy · 2006-09-27 10:10 · Score: 1

I think the important thing in the announcement is not the 80-core thing, but the idea of a memory chip sandwich. What was described is attaching the chips with what would be several thousand connection points giving more than a terabyte per second aggregate bandwidth. I heard (I watched the presentation) each core would have 256 megabytes dedicated memory.

Assuming this memory could be used smartly, segregating incoherent memory spaces (it seems rather obvious the dedicated memory would not be a coherent image of the main memory, or we would do away with any gains we got from the scale), the chip could achieve huge throughput.

And, about Tom Yager saying Intel and AMD should stop inovating... Well... Diversity is one of the two main tools of natural evolution. It served us well in the 70s and 80s and, if we can have some more of it, I say it will be great.

Send in the Windows-proof architectures.

--
http://www.dieblinkenlights.com

Terascale has scary applications by iabervon · 2006-09-27 10:24 · Score: 1

They have a slide that matches successive levels of application demand with: Text, Multimedia, Video & 3D, RMS.

Okay, so I understand that AI is more compute-intensive than video. And I understand that it could be easier (tera instead of peta) if social reasoning isn't included. But really, Intel, I just don't want RMS on my computer.

Also, the jump from nanoscale to terascale may be impressive, but I don't think it'd be useful to have a transistor with a 310-million-mile-wide gate. Your device isn't going to be useful if it doesn't fit inside earth's orbit.

Y'know, it's usually the application by Colin+Smith · 2006-09-27 10:25 · Score: 1

Throwing hardware at a badly performing application is usually the wrong way to go about getting better performance. The performance gains you're going to get are usually marginal unless there's some gross configuration mistake. The biggest performance gains are made by making changes to the application. Count that as another reason to investigate free software.

--
Deleted

Re:Y'know, it's usually the application by NerveGas · 2006-09-27 15:51 · Score: 1

That is true in some situations, but not in all of them. When you look at RDBMS work, and are talking about handling thousands of transactions per second on many-gigabyte data sets - which can't be pre-cached, and involve tying together a great deal of data sources in rather complex ways, it's very demanding on hardware. Developers of most RDBMS systems go to pretty good lengths to tweak out all of the performance they can. (Whether developers who write the SQL do is another matter.) In that situation, though, it's usually not the CPU that matters - it's the bandwidth. The beauty of the Opteron is that every time you add another CPU, you add another 128-bit memory interface, like the "big iron" provided well before AMD brought it to a commodity level.

There are plenty of mathematical simulations where memory latency is the key - no matter how much you tweak the software, the interdependencies of the various elements mean that you trample all over your memory sets. That's a hardware limitation, not application.

Even when you get into CPU-bound operation, there are plenty of instances where the hardware makes the difference. Video encoding is still the classic example where you just can't have enough horsepower.

steve

--
Oh, you're not stuck, you're just unable to let go of the onion rings.

Uselss Core War by enrevanche · 2006-09-27 10:26 · Score: 1

I think Intel is trying to sucker AMD into wasting tons of time and money trying to compete in a useless core war. Intel's architecture is behind and it needs to catch up. Intel can probably invest in two parallel development efforts easier than AMD due to it's larger size. If AMD takes the bait, it may be a big mistake. This is similar to the GHz war.

Before this will work by Solr_Flare · 2006-09-27 11:28 · Score: 1

Intel(and AMD for that matter) need to design some sort of application layer that handles parcing out tasks to the various cores regardless of the number of them. The biggest problem with multi-core applications right now is many many programs simply don't take multiple cores into account. In addition, this is going to become a huge hassle for future programmers unless this is done: "Well how many cores are we going to write this program to take advantage of?".

Also, this is something that intel/amd are definitely going to have to do on their own. They simply can't leave it up to the operating system makers to create this. I mean, look at Microsoft's 64bit Windows Vista for a great example for why you can't leave it to the OS people to do.

Until something like this is done, I doubt you'll see much enthusiasm beyond dual and quad core processors because it will take too much effort to tailor software for x number of cores with x changing on a monthly basis.

--
You are who you are, let no one tell you different. But, never close your mind to a new point of view.

Re:Before this will work by tygerstripes · 2006-09-27 20:19 · Score: 1

I'm not sure what an application layer could do for multi-threading applications.
If a program has been written to perform tasks sequentially, there's nothing an app-layer could do to multi-thread it. If a program has been written to multi-thread, then the app-layer is a useless intermediary that will just make it marginally slower.
Until there is a massive culture-shift in programming, multicore processors will simply allow you to run more than one program simultaneously (excepting specialist apps). That shift is starting to happen, but it's generational; it tends to happen as a result of the older generation moving out of the business, and the younger generation - who have grown up with the concept - moving in. The same thing happened with OOP, with website development and so on. It will happen, but visionaries who're ahead of their time are few and far between. All they can do is push back the boundaries and talk about it, so the incoming young programmers can learn about it from the start and implement it into their programming philosophy.

--
Meta will eat itself

Why doesnt the DUPE tag work anymore? by sr180 · 2006-09-27 11:39 · Score: 1

Why doesnt the dupe tag work? Someone has obviously got around it by using dup. I liked the dupe tag on an article, it meant i wouldnt have to look at 5000 comments all going "omg! its a dupe!" and "thanks to this processor we'll see this article another 78 times!"

--
In Soviet Russia the insensitive clod is YOU!

who needs 80 core cpu ? (Not apolo spacecraft) by ZXSpectrum42 · 2006-09-27 11:47 · Score: 2

from http://www.informationweek.com/shared/printableArt icle.jhtml?articleID=191901844 "I've always been amazed at the Apollo spacecraft guidance system, built by the MIT Instrumentation Lab. In 1969, this software got Apollo 11 to the moon, detached the lunar module, landed it on the moon's surface, and brought three astronauts home. It had to function on the tiny amount of memory available in the onboard Raytheon computer--it carried 8 Kbytes, not enough for a printer driver these days. And there wouldn't be time to reboot in case of system failure when the craft made re-entry. It's just as well Windows wasn't available for the job. The Apollo guidance system probably seems like routine software to technology sophisticates. Far more complex navigational systems are in operation today. The system's essentials were a few well-known algorithms based on proven logic. But to me, it's still rocket science. Great software dazzles us by virtue of what it does correctly in the face of everything that could go wrong." Whow, can you do that without 80 cores?

--
2+2 = 5 (for very large values of 2)

Not a good way to speed up general purpose apps by mpaque · 2006-09-27 12:22 · Score: 2, Informative

I like the idea of an 80 core processor. Multithreaded applications will work better

Multithreading models from the Windows/Unix/Linux community all assume equal access to system resources such as memory across all threads. They like Uniform Memory Architecture models.

An 80 core system can't really provide a uniform memory access model, as it runs into severe switching and coherency problems. (You want to snoop HOW MANY L1 caches?!??). Fancy interconnects like hyperchannel and Monte Carlo stochastic schemes start getting pinched for bandwidth around 8 cores. With this many cores, you'll wind up with computing meshes of local processors and memory interconnected using some interesting switching scheme. The article even mentions this, with a bit of hand-waving over the issues of bandwidth in shared system resources. "Intel's answer is to attach 256 Mbit of SRAM directly to EACH core. " Interconnect topology is left at a simple tiling scheme, but they are exploring ring topologies.

The result looks remarkably like a transputer mesh. I've programmed these in the past, and the model is rather different than simple multithreading. Being able to decompose the programming problem into a number of independent steps with relatively low communications demands is essential. The ability to reconfigure the interconnect topology to match the problem's data flow is essential to being able to get as much out of the processor set as possible. Without this, one can wind up with lots of idle processors, blocked on data starvation.

Comment removed by account_deleted · 2006-09-27 12:23 · Score: 1

Comment removed based on user account deletion

Cost and AMD by sethwm2 · 2006-09-27 13:07 · Score: 1

I have a feeling that AMD is going to come out of nowhere and make something that Intel will not be able to beat. AMD has been quite... too quite. Also AMD is very carful about how there priceing is set.. Intel is gooing to be big $$$ GO AMD!!! 80 cores... Make optical CPUs

A rumor of Mac OS X "thread farming" by PapayaSF · 2006-09-27 16:40 · Score: 1

Back in April, the semi-reliable (rumor-wise and server-wise) Mac OS Rumors claimed that 10.5 "Leopard" would have some pretty cool "thread farming" technology. I'll quote the whole page:

A critical component of not only Mac OS X 10.5 "Leopard," but also the Cocoa/Carbon for Windows package (more details in linked article above) will be new code co-developed with Intel that helps break up tasks into multiple threads -- therefore achieving considerably better efficiency on the next generation of multi-core Intel processors. The results we've seen on systems with up to 16 cores of Intel's next-generation "Conroe" desktop CPU architecture were amazing....with 10.4.6 as-is, the first core bears the vast majority of the workload particularly when only one or two resource-intensive apps are running.

Even when lots of different applications, many of them efficiently multi-threaded, are run on 10.4.5 or 10.4.6 only the first two CPUs are used efficiently while the third and fourth are getting plenty of work....but aren't quite living up to their full potential. Each added core after four seems to drop off in efficiency....not because OS X doesn't handle lots of processors properly, it does. In fact it's an industry leader in terms of being ready for the next generation of multicore, multiprocessor technology. It has been since day one and Apple has consistently kept it at the leading edge since then.

The problem is, simply, getting all of those core to have the maximum possible positive effect on the performance of each application. When simulating the realistic workloads of almost every kind of user, more than four cores rapidly lost any effect because there just weren't enough threads, efficiently enough balanced, to make good use of more CPU's.

Leopard changes this in every way that Apple and Intel have been able to devise. The techniques employed include tricks that both companies have been holding at ready for years, and some new things that have been developed in the past year or so to specifically address the way the "Core" (Yonah, Merom and Napa-Merom) and Codename 'Conroe' architectures work. Most of it goes beyond our technical competency; we're sure that the folks at Ars Technica will have a lot to say about this in the next few months as more details leak about the hardware and software involved in these enhancements.

Some, but certainly not all, of these techniques will eventually make their way into Intel's optimized in-house compilers. Some will even become part of the GCC compilers that are critical to building OS X and indeed most Xcode applications, eventually. But right now they are by and large highly experimental, being part of an operating system codebase that is not even quite "alpha" in terms of usability.

That said, it's a thing of beauty to see 16 cores used with bizarrely perfect symmetry even when performing relatively simple tasks that have nearly no application-level threading in their collective codebases. 32 cores work nearly as well, and somehow manage to make tasks that would normally only max out one or two cores and be unable to go beyond that point, spread out across nearly all the CPU's with a beautiful cascade effect created for just such a demonstration in the Leopard version of Activity Monitor (just wait until you see all the 3D OpenGL visualizations that have been whippped up....but that's another article entirely and bordering on embargoed territory to boot!).

--
Q: What does the "B." in Benoit B. Mandelbrot stand for? A: Benoit B. Mandelbrot

Re:A rumor of Mac OS X "thread farming" by jericho4.0 · 2006-09-27 17:55 · Score: 1

Interesting. Delivering this could be a huge boon for Apple, as I don't see MS and the Vista codebase as being very agile, and multicore is all Intel and AMD will be talking about for a long time.
I'm also looking forward to seeing dtrace in Leopard. I think I might be buying a Mac!

--
"A language that doesn't affect the way you think about programming, is not worth knowing" - Alan Perlis

Please take a look at OpenMP.. by chaboud · 2006-09-27 17:53 · Score: 1

Let's say I have a for loop. These are pretty common in code (obscenely common wouldn't be a gross overstatement).

for( ii = 0; ii < NumberOfElementsInBlah; ++ii) { Blah.DoSomethingUniquetoThisElement(ii); }

So, let's make this work for multiple processors, scaling to the system that we're running on:

#pragma omp parallel { #pragma omp for schedule(static) for( ii = 0; ii < NumberOfElementsInBlah; ++ii) { Blah.DoSomethingUniquetoThisElement(ii); } }

What do you know?! We're done...

Now, of course, there are data dependency and variable interaction problems to handle (which OpenMP provides facilities for), but, if Blah.DoSomethingUniquetoThisElement() were actually your work, you'd be done with this code. It's really that easy.

RTFA by Don_dumb · 2006-09-27 19:06 · Score: 1

The problem is not the number of processors but actually making use of them, so you *can* have enough cores. There is no point having an 80 core system if ony two of the cores are being used.
What Tom Yager is saying is that AMD & Intel makers shouldn't just fall into a race to see how many cores they can fit on a chip but actually getting architecture and software that ensures that adding new cores does actually give a performance advantage.

--
If this were really happening, what would you think?

Re: Matlab in Parallel by Anonymous Coward · 2006-09-27 22:11 · Score: 0

is MRI for Magneto Rotational?

I have seen PMatlab, MatlabMPI paper recently ( http://www.ll.mit.edu/MatlabMPI/ )
Though have no time to implement/check it ...

80? Call me old fashioned.... by maximthemagnificent · 2006-09-28 00:04 · Score: 1

Maybe I'm a throwback, but I'm still more interested in speeding up a single thread than in having 80
seperate ones. It's fundamentally harder, but that's why it's useful, no?

Connection Machine, MassPar, Tera .... by peter303 · 2006-09-28 02:06 · Score: 1

There have been several super computer companies who have come and gone on the premise of fine-grain parallel processing, such as mentioned in the title. All of them used some version of UNIX (CM had a pre-UNIX OS at one time). The reason for this was that UNIX was the "Linux" of the 1980s and early 1990s: low cost source code license, porting experience to many machines etc. They all had OS-extensions and C-language constructs for managing fine-grain parallelism. So my point is there is a lot of experience out there int his area.
These companies died due to using custom hardware they could upgrade only in 3-5 year generations. Clustered commodity workstation/PC CPUs generally upgraded 3-5 times faster and "caught up" in price/performance.
If a cell-processor can emulate the x86 instruction set fairly transparently, then they could finally beat this fine-grained jinx.

Scared 4 Apple by aJester · 2006-09-28 03:52 · Score: 1

I am worried for Apple.
I hope they don't get suckered into this wild Intel adventure...

I hope Steve Jobs doesn't buy into this and instead focusses on getting Mac OS X to scale well on 2 - 4 cores.. (perhaps 8 cores for OS X Server).
IMHO, a normal consumer should NOT need more than 2 cores.
And they do, it indicates that there is something wrong with the 2 cores that are there ...Or the OS running on it.

I somehow am *very* cynical about new throwing a few dozen cores in the chip trend.

Just-a-Jester

220 comments