Slashdot Mirror
Intel And AMD's Dual-Core CPUs Investigated
Hack Jandy writes "Anandtech has a bunch of insider information concerning Intel and AMD's move to dual-core CPUs. The article has lots of great information on how the move to dual-core processors affects modern computing - in particular, Anand sees more promise in multiple CPU cores that perform different operations, rather than just stamping two identical cores on the same processor like AMD and Intel are doing now."
Look Ma! I got a Ferrari that when you press a button becomes a Yugo!
/^[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,4}$/i
I would rather have faster processors than multiple cores, as it is not enough is multi-threaded. Even the highest end 3D apps, their render engines are SMP capable, but all geometry translation/deformation is not. That would be one core right? Unless multiple cores could show up as one single core/proc in the OS..
The applications simply aren't there, as AnandTech mentions. Hyperthreading, for instance, did not cause sudden and dramatic speed improvements. The only benefits we're going to see are with applications specifically written for multiprocessor systems. These can take full advantage of the strengths of dual core CPUs.
US businesses that currently accept chip and PIN/signature
And let's call it, say, the x87 math co-processor? :)
Naw, you really need two of the same chips in there. Too much steering of processes and whatnot otherwise.
-m
http://www.invisik.com
Its seems that Intel have lost their technology edge. Early in Intel's life, the company direction was driven by the engineers, but it over the last few years, highlighted by the mhz race, all tech R&D has been driven by marketing managers. This was probably to be expected. Marketers and non-tech managers are usually very good with people, very good at playing politics, and hence very good at influencing company direction; far better than most engineers. Intel is now paying the price for their incompetence by loosing out to smaller, more hungry competitors.
0 (Project Niagra). Intel certainly has much catching up to do, but its time for a new race and hopefully they'll get their arse into gear and show us some exciting things in the years to come, that is, if the marketoids can be somehow dethroned from their positions of power.
I don't know where the Itanic fits into this theory. I guess if it wasn't so late, and was made available during the tech bubble, Intel would now be on a fundamentally different track, rather than playing catch-up (poorly) with more innovative companies.
Now, onto multi-core chips. This is actually a very exciting direction. Sun has already demonstrated an 8 core, quad-hyperthreading 32-way chip http://blogs.sun.com/roller/page/jonathan/2004091
Moving processing out into special purpose processers, and then back into the main one again as Moore's Law takes effect has been known about since the term the wheel of reincarnation was coined back in 1968.
I think they should dynamicly change the clock speed based on heat content. Have a max hz, then have it slow down the hotter it gets. Then you could remove the cpu fan and not worry about it, save the fact that it would be slow as dirt.
I think the CPUs would be the same speed sorta. Just have one tweaked for say floats and the other something else. If you have a float heavy process you use core 0 and otherwise core 1. You can end up with the same CPI for standard loads but with some programs would do better with one than the other, as they aren't standard.
It is no longer uncommon to be uncommon.
What is being referred to here is the possibility of having different cores, not just two identical cores on the same silicon. Similarly to how the PowerPC970 has two different branch prediction algorithms which "compete": each calculating which branches should be taken, with a central heuristic keeping track of how well each has been doing lately and chosing which will be used for the next series of branch predictions, a heterogeneously cored chip could offer several differing implementations of the same realestate. This could mean having one core with 4 FPU's/2 IU's and another with the reverse, or different length pipelines/branch predictors/L1 caches - thus opening up the possibility of CPU hierarchies, where set A is really good at certain tasks and set B is really good at another, and the OS is smart enough to schedule them appropriately. Think of a machine which is used for both compilations and running jobs, or think of the benefits in a virtual machine environment! The admin could partition the system along functional boundaries (intelligent hyperthreading).
Another possibility is where the entire system is devoted to a single task (think HPC: fluid flow, weather simulations, etc) where you could have threads doing the intensive floating point calculations on one core, and the heavy integer arithmetic on the other, or maybe split up the cores based on memory accesses patterns, or cache use, or built-in ASICs!
What I would love to see is a system where you have 2/4 cores with a large cache, plus an FPGA or two on die that each application can program - with OS cooperation this could be a "killer app" in silicon. Do a lot of "int*float*sqrt(int)?" - then program the FPGA to do it in one operation, as if the original chip design had it all along!
Insanely cool stuff! "CPU and GPU", sheesh.
I can't fucking wait.
Is just me or does it seem odd they are using x20, x30, and x40 for names? I guess x20 + x30 + x40 does make an x90, slightly better than my x86.
Yep, this would be cool. However it sort of already exists, but coming from the other angle. I know at least xilinx has a powerpc builtin in a number of their high end FPGAs. But your approach could possibly very interesting for homeusers. A more high-end cpu with low-end FPGA.. Guess you just have to invent the first killer-app that would use it :)
Processor: shall be defined as all processors where the Oracle programs are installed and/or running. Programs licensed on a Processor basis may be accessed by your internal users (including agents and contractors) and by third party users. For the purposes of counting the number of processors which require licensing, a multicore chip with "n" processor cores shall be counted as "n" processors.
This is from Oracle's "Licensing Definitions Document," the emphasis is mine. I found it on the partner web site, which I'm pretty sure is inaccessible to the general public.Of course, I expect this to change (esp. on Windows) p.d.q. given Microsoft's recent announcement.
My old PC had this, it was called a turbo button.
-Copyright law #69:Whenever Mickey Mouse is about to enter the public domain,copyrights get extended by 25 years.
At least yesterday they were still in.
Amds dual core chips dont use a local HT link to for core-core communication. They have both cores linked to a crossbar, which also has ports for the HT-links and the memory controller.
So a dual core chip still has 3 outgoing ht links, allowing to use 8 dual core chips in one system without "glue"
HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
In terms of "marketing speak", this is a good opportunity for Sparc and PowerPC chips to catch up to the X86 architecture.
Thanks to Intel's own marketing, most users are used to seeing that Mhz = power, and Apple suffers from the fact that the G5 tops out at 2.5Ghz, while Intel chips cruise along at 3+Ghz. Sun's SPARC architecture suffers from the same illusion, although comparably, both the Sparc and PPC architectures are quite close to X86 in terms of actual horsepower (not so much with Sparc, but Sun's true power is total throughput and reliablity and scalability, not flops).
With Intel "stuck" at around 4Ghz, IBM/Apple could figure out how to ramp up the G5 (or it's successor) to 4+Ghz, and beat Intel at it's own marketing game.
Similarly, this bump in the roadmap for Intel could be the opportunity for other/alternative CPU architectures to gain some marketshare.
(Posted as someone very, very tired of the Wintel Monopoly)
If telephones are outlawed, then only outlaws will have telephones.
It looks like IBM chose the right direction to go with their line of processors. With things like the power 5 chip, and altivec processing units combined you get more bang for the buck vs a dual core x86 chipset running at a higher clock speed.
However I dont see a mass migration to the power platform due to the entrenchment of the desktop market. BUT if they can proove they have the more powerful upgrade path we may be seeing more powerPC type servers in the farms as businesses upgrade and look for that power for price. With PPC linux this will be possible and Microsoft will be sitting around wondering what the hell happened.
There is a very interesting article in the last edition of Fortune. I think AMD got it right this time around.
:
s /0,15114,724543,00.html
:(
My favorite quote
AMD CFO Rivet explains
"As hard as we tried to win the hearts and minds of CIOs, with the desktop as our focus we were going to fail. They made their decisions with the server on down. When Intel had 100% of the x86 server market, it could charge whatever it wanted and use that money to beat us on desktops. We had to be in the profit haven".
Ruiz (CEO of AMD) calls the server-led approach "do or die" for AMD: "If we hadn't pulled this off I would have shut the door"
From the Fortune article:
AMD: Chipping Away at Intel
CEO Hector Ruiz came from humble roots to propel AMD into the big leagues.
http://www.fortune.com/fortune/technology/article
You need to be a subscriber to read the whole article
A hungry bear does not dance!
On a single CPU system, the X client and server compete for time. It can sometimes be faster to run certain apps over a fast network than locally on the same machine.
On a dual machine or multi-core machine the client and server can both be given time on separate CPUs or presumably different cores on the one CPU.
Deleted
Why is this better than current implementations of multiple processors on one motherboard? Isn't it fundamentally the same thing, just using less space? Granted, using less space allows for more and more processors to be placed in the same area, just like making smaller transistors allows you to place more of them on the chip. But that's only really useful in a server environment, which is why it never really caught on among home users. Will this really provide any benefit for everyday users?
I wonder if we have reached the end of the race for processing power. Up until the 60s and 70s, car manufacturers were trying to create increasingly more powerful engines. In part because of the gas shortages of the 70s, but also due to the fact that people really can't do anything useful with 700HP, other than kill themselves really really fast. The focus shifted from power to economy. Maybe one day manufacturers will be touting that their laptops are so efficient they can power themselves from the kinetic energy of moving them around, like watches of today. This will be a sad day for most of us, because it means we will have to get off our asses every once in a while to keep our laptops running.
On a side note, how come nobody's posted anything about a Beowulf cluster of these?
--
https://www.eff.org/https-everywhere
Certainly about how AMD do dual-core, which as it has been detailed since 2001 (and talked about since 1999) I think is extremely poor for a large website like Anandtech to get wrong.
See comments 50, 51 and 54 that go with the story to see how AMD actually do dual-core (they don't 'fuse' hypertransport links together, like the article says they do)
What is sadder is that they haven't corrected the story even though the incorrectness has been pointed out to them in the feedback, and presumably via e-mail as well. Nothing in the article can be trusted in any way because if basic facts are ignored, then what about the rest?
I certainly do not think that such poor articles should be linked from Slashdot. Why should AnandTech get rewarded for such shoddy work?
Maybe we'll see dual-core CPUs where the second core does some
of the 3D-calculation today's graphics chipsets do?
That would certainly be useful for some fields of math.
I think Anand was suggesting that in his article. While the schedulers of Linux and some of the other OSes may be able to handle that, I don't think you want to go that way given the hacks that are used in schedulers, e.g. the hack that Linux uses when running a high priority and a low priority thread on the same hyperthreaded processor. All system accounting is done in terms of processor run time and on an ASMP system, run times aren't going to be equal.
This is a tricky time for hardware manufacturers - how to promote upgrades which are essentially placeholders for a new hardware generation - and hope like hell that Microsoft will actually promote applications that will use that new functionality. Because Microsoft can afford to lose their R&D money, Intel and AMD cannot.
Don't get me wrong, I'm looking forward to true 64-bit dual core architectures on the PC platform, but unless something amazing happens in the next 12 months, Microsoft will again be the gatekeeper to the mass uptake of that hardware, geek rage and linux notwithstanding. The shark will get it's DRM when the makers are appropriately terrified, and even then they may not make their money back.
From a manufacturer/reseller point of view, it's not looking all that certain. Uncertainty is deadly to the CPU/mainboard market, and I'm seeing it in the hedged bets of computer swapmeets and resellers. The explosion of mp3 players, digital cameras, dvd burners and the astonshing fall in solid state memory might take up the slack for now, but that still means those crucial early-adopters aren't looking at the new goods.
We live in interesting times.
insecurity asks the wrong question irritation gives the wrong answer
The only processor where your claim is true, is the 486SX, which had indeed the floating point unit disabled. When you bought the 487 (or Overdrive, not sure there), it was essentially a 486DX processor which turned off the 486SX processor.
The joke on the customer here was that SX and DX means something completely different on the 386 chip. (bus speed was doubled on a 386DX and not on a 386SX)
Ahhh...the great dumpster continuum. Many a free computer will be found there. -- sowth (748135)
In general purpose computing it would be nice to have one core dedicated to mathematically intensive tasks and one for the housekeeping. So that while you compile your X does not hang.
My Aurora : http://www.youtube.com/watch?v=o91ZsGwJYyg
FB : https://www.facebook.com/TanveersPhotography
Would it be possible to have a dual core processor with both a PPC and a x86 core?
There was interesting paper at ISCA a few years back that proposed vector extensions to the Alpha ISA (called Taranula) and then making a dual core processor with the second core a vector core. The vector core would still be dependent on the scalar core for certain functionality (eg, supplying scalar arguments, renaming) and they proposed a 16MB!! L2 cache to feed the beast, but the performance numbers (especially the performance/power numbers) were pretty impressive.
What is being referred to here is the possibility of having different cores, not just two identical cores on the same silicon. ...
Another possibility is where the entire system is devoted to a single task (think HPC: fluid flow, weather simulations, etc) where you could have threads doing the intensive floating point calculations on one core, and the heavy integer arithmetic on the other, or maybe split up the cores based on memory accesses patterns, or cache use, or built-in ASICs!
The problem with this is that by designing resources to be used in "certain situations" you put resources in the mix that may not be usable most of the time. Using your example of having one core with 4 Integer units and 2 FPUs and another core with 2 Integer units and 4 FPUs, why not make a single core with 6 Integer units and 6 FPUs and let the core itself how to allocate those resources based on runtime requirements (as opposed to potentially having one of those cores idle until a "special" program is run)? In fact, why not just have two cores with the average of each (3 Integer and 3 FPUs) so that complex scheduling doesn't have to be written (and it would be complex... you'd have to know before the thread is scheduled which core would be more suited to it, either through analysis of the executable code or through hints given at compile time).
I guess that one of the things is that due to the widely varying usage patters on consumer PCs vs. complexity, you'd be better off making a compromise and designing based on average usages. If you are designing specifically for systems that will be heavy in CFD or the like, design a CPU that is an FPU beast. These two markets have pretty different requirements and by trying to blend them, you'll end up with something not really suited for either.
The other thing is that when you design specific silicon (different types of cores), you have to make sure that there's sufficient usage of each to justify the effort required. Having two cores where one is idle 99% of the time is a wasted effort.
The thing that irks me is that there is no general computing source any more. Things have pretty much descended into the various "camps" with pee cee people reading about those new processors and the Mac people reading about the Power PC processor.
I used to be able to keep up with processor design in Byte Magazine. It also kept me apprised of each different computer that came out back when no one computer type and operating system had over 90% of the market and I think that Byte helped serve those who didn't want to see Microsoft-Intel become as dominant as they have become.
The death of Byte is still a sore spot with me. I ran an Intel platofrm for many years and was able to keep up with what Motorola and Sun were doing with their designs. There were even columns on embedded applications. I felt like I had a really good handle on the microprocessor universe and the differences. Sadly, not so now (or should I use Jerry Pournelle's frequent "Alas...").
Gods don't kill people, people with gods kill people.
Intel's P4 can't go any faster because of heat, and they can't do anything about it.
The hell they can't. Three words would fix Intel's heat situation easily: Desktop Pentium M. Where can I buy such a motherboard?
I have a dual Athlon MP 1200 board, and before that, an Abit BP6 (dual celeron). There are advantages to having dual CPUs. One of them is, if a rogue process suddenly starts using up an entire processor (a situation that would bring single cpu systems to a hard-lock) you might not even notice a performance problem until you try and use that process. You can run twice as many processes and won't see a performance hit (provided you have the RAM). For example: I can run about 4 instances of Diablo II Expansion, Firefox with about 10 pages open, and tons of other little things in the background. I'm currently running 46 processes, including 3 diablos, Firefox with 7 pages open, AIM, Rapidbackup, Google desktop search, gmail notifier, getright, Ultraedit, TrayIt, Windows Sniper, Clipomatic, Transtext, Tclock, stickies, powermenu, winbar and all the usual system processes. This is the normal state of windows for me and it runs just fine.
However there are disadvantages too. Good luck finding a soundard with lots of features that gets along with dual CPUs. Creative has awful drivers and I'd almost swear they don't bother testing them, most other soundcards do just as bad or worse and offer fewer features. I built this machine back in fall of '01 and it wasn't until about a year ago that they released a set of drivers for the Audigy that I couldn't cause a BSOD at will with. If I ran Winamp using the directsound out and seeked around within a song repeatidly really fast it would BSOD 100% of the time. Not to mention you have to buy TWO processors rather than one, and the board was ~$500, is E-ATX, barely fits in an Antec SX1200 (HUGE case). In fact the hds stick out over the DIM slots and almost over the 2nd CPU. My case is gigantic and its too small for this motherboard.
Question everything
There were some wierd Mac variations in the 1980s with a second CPU on a plug-in board. They could run Photoshop faster, but otherwise were useless.
There are really only two multi-CPU architectures that are generally useful: shared-memory symmetrical multiprocessors, and networked clusters with no shared memory. Many other architectures have been tried - partially shared memory machines, shared-memory machines where some CPUs lacked some features like floating point, hypercubes, single-instruction-multiple-datastream machines, and dataflow processors. None has achieved lasting success.
About the only unusual architecture ever sold in volume is the Playstation 2, with two vector processors. Even there, the vector processors are mostly used as a GPU. (Although one major game physics engine actually runs in the PS2 vector processors, an impressive achievement.)
Programming for wierd architectures is hard, requires much tool development, and results in programs tied to specific hardware. So it doesn't happen much. That's why the wierd architectures fail. They're never that much faster, and by the time the software works, the hardware market is somewhere else.