People question IBM and Intel's "quadcore" CPUs
Both Intel and IBM's "quadcore" are actually double dual core: a MCM (multi chip module) that glues two dual core CPUs together. Only Advanced Micro Devices and Sun Microsystems have the ability to design sophisticated true multi-core CPUs.
The above are people's opinion.
18 Comments:
The truth of the matter is both Intel and IBM could (read: most likely will) design quad core chips. In the grand scheme of semi-conductor manufacturing, dual core is still in it's infancy in terms of adoption which makes quad core chips more like the in the stages of invetro fertilization. Crudely speaking in those terms, AMD is ahead of the game because they were the first to jack off into the test tube.
While I'm very interested to see how the quad core -vs- double dual core race will pan out, I'm very inclined to be worried about AMD's manufacturability. Their design takes up an awful lot of silicon real estate to make on chip. A defect or any sort of particulate on one chip would seriously render 1 or more cores useless, in which case the entire die (or die field assuming AMD's chip is a single chip/per pass lithography design) would be garbage and unfit for sale.
The counter point is Intel's double dual core. The advantage they will enjoy in terms of defects affecting yield is a one core loss only kills at most 2 die sizes worth of silicon. Additionally and probably better in terms of cost and performance, each dual core of the double dual methodology could be optimized by being matched to another dual core chip based on how each bins out in sort. As we all know, chips on any wafer size bin out faster in different regions of the wafer. Intel's design will more than likely allow them to enjoy higher yields and final components with boutique lines of better processors to bring to the market place.
Either way... I look forward to seeing how both companies fair with their designs. Both are great innovators no matter which way you look at it.
Cheers, my friend (and often adversary)!
The only true quad core is one on one peice of silicon. That Pentium D crap is a joke. That was never true dual core. And the quad cores will be the same, fake. Its only as good as 2 dual cores in a dual socket for what its worth. Only AMD's had true dual cores.
AMD is the only true quad core! Nobody can say 2 dies glued together is a true quad core because its simply physically not. Its 2 cpu's on one peice of cpu board. Only one die on one cpu board is true quad core. No exceptions!
For many of today's applications, two dual-core processors may be more effective than one quad-core processor.
There is very little today in common applications that scale to more than 2 threads.
So if these two threads are programmed to have affinity on one shared cache dual-core chip (i.e. one Conroe/Woodcrest), these threads will perform well.
Depending on how well the various affinity binding capabilities in the OS and in apps are used, Intel/IBM's approach will do just fine.
Yes, two dual-core chips will not do quite as well for applications that have achieved 4 thread parallelization. But maybe such apps won't be common until 2010+.
It may be that AMD is ahead of the curve 2-3 years and while being the technical leader, will not derive any great business benefits from being ahead of the curve.
What is driving AMD today is its low cost of production. And that is what Intel is trying to match by shedding non-core business units and cutting back on staff.
Good Lord, does it really matter? One gets four cores in a package. I suppose if two dual cores are hard wired together, there will be some latency. So what, it's still a huge advance in computing. It seems that grown, educated adults have a lack of minutia to argue over.
"For many of today's applications, two dual-core processors may be more effective than one quad-core processor."
If applications scales to two CPU's there is little point in not making it to scale to N CPUs. With games it is a bit more difficult since trually SMP capable games haven't been developed for all that long and usually they simply separate discrete parts to compute on different cores. If they would divide computing-heavy things to multiple threads things would scale way better. Only problem is that this is much harder to code than.
"There is very little today in common applications that scale to more than 2 threads."
I'm quite sure most, if not all, rendering, archiving and audio/video encoding programs that do SMP can do it on N CPU's.
"So if these two threads are programmed to have affinity on one shared cache dual-core chip ..."
"Depending on how well the various affinity binding capabilities in the OS and in apps are used, Intel/IBM's approach will do just fine."
Problem is that good thread schedulers are hard to write. XP one sucks bigtime, I don't know about Vista. Linux has rather good one that can differate between real CPU's and virtual ones (Intel HT) and divide work accordingly. Also there are special schedulers for multi-socket and multicore CPU's.
"Yes, two dual-core chips will not do quite as well for applications that have achieved 4 thread parallelization"
It's highly dependant on application. Lots of things can do just fine with dual-chip CPU.
Things that can't scale that well with such configurations won't scale that well with on-chip connected multicores too. Only difference is the efficiency coeficent. They are both rather bad compared to linear scaling, only that the non-connected one has worse coeficent.
Seems like a smart plan by Intel.
"ONLY Advanced Micro Devices and Sun Microsystems have the ability to design sophisticated true multi-core CPUs."
Sun launches 8-core Niagara
(32 simultaneous threads, 64 bit)
Azul launches 48-core processor
(821-million-transistor chip, 64 bit)
IBM, Sony and Toshiba CELL processor
(Well... it have 8 calculating units)
Fujitsu FR1000
(4 FR550 32-bit cores capable of executing up to eight instructions simultaneously)
intellaSys SEAforth-24
(The processor features 24 cores, each executing one high-level Forth instruction per nanosecond, and needs only 150 mW in a typical application)
"Analysts Dean McCarron and Nathan Brookwood say multicore processor designs that utilize MCMs are no less "true" multicore offerings than the monolithic, or "native," designs from AMD."
..
http://www.xtremesystems.org/forums/showthread.php?t=103982&page=3
http://www.iamxtreme.net/coolaler/conroe/KENSFIELD/8.gif
http://www.iamxtreme.net/coolaler/meromvsconroe/c09.gif
In tmpGenc Kentfield at 2.4ghz is 50% faster than an X6800 Core2 at 2.93ghz. How true do we need:D
besides things explained in article in informationweek,
there are others.
1st its production cost. die size of native quad core is
more than 2 sizes of dual-core. this means that possibility
of broken die is 2 times higher. even if intel yields
are lower than amd's as u think, native quad core destroy
this difference.
however, everything explained in informationweek is not
important. important thing is that amd cannot use glue
approach at all, even if they will want to! they cannot
because of IMC! they use IMC benefits such as extra
low memory access latency, but pay price as well, which
is inability to use glue approach. taking 2 amd die
together means that second die has to access memory via
1st die and HT link (there is only one in usual desktop
amd dies, thus special dies needed). second die will have
very slow memory access speed and terrible latency, coupled
with small cache it will show zero benefit of 3rd and 4th
cores! amd is simply forced to use native quad approach.
while intel can play with it, introducing native die
whenever it will be good from manufacturing pov
(45 for quad and 65 for dual).
sun example is not equal, because sun cores do not equal to amd
and intel ones. sun cores share 1 fpu unit for all and so
on, while amd/intel cores are fully functional units.
"Yes, two dual-core chips will not do quite as well for applications that have achieved 4 thread parallelization. But maybe such apps won't be common until 2010+."
besides application itself there is os/driver level which is also doing a lot of work. GUI/3d, tcp/ip stack, disk IO.
apple demostrated on last wwdc 2 weeks ago MT optimized open gl layer, which alone doubled frame rate of world of warcraft! this is only one simple example.
Software speaking Intel, IBM are true dual core, quad core, ...
Manufacturing/technology speaking isn’t true dual core, quad core.
Point is doesn’t matter if it work, however the all idea that it stays, is that if there was more space in the package, we would see 8, 16, 32 cores from Intel and IBM.
But final point is what does make them 2 cores, 4 core, software detection or how it was manufactured?
But final point is what does make them 2 cores, 4 core, software detection or how it was manufactured?
I believe final point is how well those cpus scale with the number of cores, that's why IBM's and Intel's approaches are being critizised.
But final point is what does make them 2 cores, 4 core, software detection or how it was manufactured?
Exactly! You give the users and software developers a 4 cores in a chip, which put into the single socket, and they enjoy the performance of 4 core/threads.
put it in the simpler perspective of a dual core scenario. When intel first come out with the dualcore, it is the time that user get enhaced responsiveness and developer get 2 thread running concurrently.
Refer to the recent released IBM and Intel quad-core To the users, the software developer, the industry, it is indeed quad core, to the supporters belonging the other camp, it just simply NO-NO without a single thought :)
"apple demostrated on last wwdc 2 weeks ago MT optimized open gl layer, which alone doubled frame rate of world of warcraft! this is only one simple example."
It is a good example of which there will be MANY with Apple's OS. One must remember that most of Apple's OS is a monolithic chunk of mono-threaded code. Many operations in OS X use 1 of 2 global locks and end up locking half the OS.
True, Apple has done a lot of work on Mac OS. But it will take 5-10 more years to make the whole thing authentically and pervasively multi-threaded.
Apple wins with Objective-C, but also loses. There is no real support for Objective-C outside of Apple, so Apple has to do all the work of making this compiler, including all the tough threading stuff. This is one factor that contributes to how long it is taking Apple to improve their OS.
I really like my Mac, but fast it ain't.
The fabric needed to stitch together 4 cores on the same chip piece of silicon is a monstrous piece of work. At the same time, the effort required to put together 4 dual core chips with dedicated L3 is no small feat, either.
The name of the game is performance vs cost. If you can get close quad core performance for the cost of gluing two dual core chips together on the same CPU board, who is smarter? The company that took the time and money to develop a quad core chip or the company that figured out is was almost as good to glue two dual core chips on the same CPU board?
BTW, the same chief architect currently leading the way at AMD moved IBM server chips to dual-cores before they were in style.
Well, at least IBM is shipping out tons of three-core chips inside Xbox360, so it's not that they don't know how to put more that two cores on one chip :-)
All these years AMD's Athlon 64's were performing amazingly against anything Intel had to offer to date. Now Intel releases there upgraded versions of there Pentium M (Pentium 3) called Core 2 Duo, which are faster clock for clock? Well, so what? AMD's current K8 designed years ago is still superior to Intel's "So Called" new Core 2 Duo architecture. This FSB business of theres is so useless that Intel is required to increase there L2 cache at an insane amount for there CPU's to hide away the badly bottlenecked FSB.
I do give credit to Intel for at least releasing something a lot better then the dogly piece of junk Pentium 4 CPU, but come on, AMD was king, and now because of the Conroe, many of you want to kick them in the balls? Give me a break.
Let Intel have a little spot light with there Conroe’s for now, O.K. because AMD’s Core 2 Duo & Kentsfield killer is coming soon.
Just wait for the "TRUE" Native Quad-Core to be released. Yes, the “First True Native Quad-Core”, and not this “Double Dual-Core CPU” that Intel is planning on releasing LOL. And if you think AMD is done, no, they have an even better CPU architecture then there Native Quad-Core, all thanks to ATI's buy out ) Just wait for 45nm, and see who takes the performance crown AGAIN?
Why is everybody killing AMD????
So, why is that now, after about 3 years AMD'64 architecture destroyed the Pentium IV with a design which is technically still superior to Intel's design, that everybody wishes to kill AMD??? AMD did something to the market Intel was not doing for AGES!!!!! They reinvented the mainstream microprocessor market, they created a piece of engineering artwork. AMD 64 design is elegant and much more intelligent than Intel's one. Of course, Intel's latest seems to be the best thing ever existed, which I don't believe it is and AMD will soon prove again to us it has a better design. Intel had to resort to TRICKS all the time to make it's processors competitive. SSE3, Hyperthreading, gigantic caches, not to mention a lot of cheating in compilers, tests and applications. I won't even mention Intel's business practices, dishonest is the least I can say about them. AMD64 would have crushed Intel's processors long ago if it wasn't for the people they paid not to sell AMD products, so, I don't want to buy stuff from a company that resorts to those methods instead of moving that money into engineering and developing capable products. The Pentium IV was a flop from start, Intel's new processor only corrects that. Where's Hypertransport??? where's the on-die memory controller and so forth??? I see a HUGE cache, now, let's move to applications that don't use cache very well. Intel took 3 years to play catch up with AMD, but AMD did not sit around, their timing is still a little behind for many reasons, but they will in less than a year demonstrate the K8L, which will surely kick intel's butt once again. The design is still superior, AMD has given us TRUE DUAL CORE and will give us TRUE QUAD CORE, not "GLUED" processors as Intel. Stop killing AMD and respect true engineering. If you have doubts about the best processor, go look at Cray's supercomputers, you won't find intels there.
Post a Comment
<< Home