backtop


Print E-mail del.icio.us 42 comment(s) - last by Gnarr.. on Jun 2 at 5:51 AM


ClearSpeed Advance board

The CSX600's design; 96 dedicated math units each with their own 6KB cache
Providing nearly 5X the math processing power of an AMD FX60 CPU, AMD may use Clearspeed's processors

While current generation processors are good at arithmetic in general, their design leaves many things to be desired. For example, while an Opteron CPU is more than capable of rendering 3D graphics, special purpose processors from ATI and NVIDIA are infinitely better at this task due to their specialized designs.  Likewise, ATI and NVIDIA GPUs would not make very good general purpose processors. Both Intel and AMD have worked very hard on making the latest generation processors very good at general purpose computations -- multiple cores, branch predictions, deep pipelines, operation fusion, etc.  Unfortunately, none of these trends are particularly necessary for heavy mathematical computations, and in many cases these advancements have proven detrimental. 

While desktops generally do not rely on heavy math operations, workstations and servers are demanding more and more out of CPUs that have drifted further and further away from mathematical computations.  Enter Clearspeed. The company has been in the talks with AMD for some time now over the use of its dedicated math co-processors. AMD is looking at Clearspeed's products and possibly integrating Clearspeed's silicon into future Opteron quad-core processors.

Clearspeed's current flagship processor, the CSX600 is a processor dedicated to complex mathematics processing only. Used in designs where applications include medical research, CAD, space research, data mining, and other math-intensive applications, Clearspeed's products are far superior to standard general purpose processors alone. Clearspeed claims that its CSX600 co-processor is able to perform at 25Gflops per second under the right conditions. AMD's Opteron can handle approximately 5.7Gflops, though the operations on the Opteron are significantly more complex than the ones carried out on the CSX600.

Currently, Clearspeed's products are available as individual co-processor units or as add-in PCIe board. The company says that while PCI Express does offer a significant speed advantage over using PCI-X or PCI interfaces, HyperTransport interconnects offer the best performance. Clearspeed also mentions that Intel's latest platform performs well with its co-processors, but due to delays with Common System Interconnect (Intel's answer to HyperTransport) performance gains are limited.

As of right now, there doesn't seem to be a firm deal between AMD and Clearspeed just yet, but things are shaping up nicely for Clearspeed, which is based in the UK. Intel also uses Clearspeed's technology in some of its line of processors.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Could help...
By TheLiberalTruth on 3/18/2006 2:50:50 AM , Rating: 2
I could see these being used in high end gaming machines to offload some physics calculations from the CPU.
It would probably be of use if you ran SETI or some other distributed computing project(anything "@Home").
Maybe you could even find a 1.0x10^7 (10 million) digit prime and win yourself $50k. Just kidding.




RE: Could help...
By Hypernova on 3/18/2006 3:34:20 AM , Rating: 1
"I could see these being used in high end gaming machines to offload some physics calculations from the CPU."
Those already exist in the form of Ageia PhysX cards.


RE: Could help...
By rgreen83 on 3/18/06, Rating: -1
RE: Could help...
By krwilsonn on 3/18/2006 8:09:58 AM , Rating: 1
quote:
Actually they dont exist, unless you would like to point me to where i can buy one.


So they don't exist since you can't buy one? Well, then, San Francisco doesn't exist because I can't see it from here. Or whatever


RE: Could help...
By bob661 on 3/18/2006 10:14:20 PM , Rating: 3
If you can't buy it, it's vaporware. The comparison with San Francisco is idiotic, at best.


RE: Could help...
By Hypernova on 3/18/2006 8:11:22 AM , Rating: 1
ASUS got cards made already, it's just that untill Ureal engine 3 games arrive it's useless so no sense mass producing and distributing them now.

GooglePic "physx" and pi'c of the finished cards are every where.


RE: Could help...
By Griswold on 3/18/2006 11:03:35 AM , Rating: 2
Thats a funny excuse. Its a niche product and will remain as such for quite some time, especially with ATI (and probably Nvidia as well behind closed doors) pushing for solutions that come with their own GPU's (and certainly not just shader units doing the calculations as announced by ATI). If there is a market for such a product, these two will take it and not leave much left for third party add-on cards.


RE: Could help...
By Hypernova on 3/18/2006 3:18:26 PM , Rating: 2
Well these things certainly ain't for general gamers I give you that, from the pic's it looks like it will take up another PCI-E x16 slot so any current boards except Quad Royals won't be able to SLI/Crossfire.


RE: Could help...
By masher2 (blog) on 3/19/2006 12:55:35 AM , Rating: 2
> "Those already exist in the form of Ageia PhysX cards."

No, Aegia is customized for "gaming physics", which is primarily multibody kinematics/dynamics, collision-detection, etc. It would be essentially useless for many types of physics calculations, whereas this chip would not be.


RE: Could help...
By Furen on 3/18/2006 3:55:00 AM , Rating: 2
I see this more as a drop-in replacement for an Opteron in multi-socket systems more than anything else. The thing could be fitted to communicate with the system with hypertransport... heck, it even has its own DDR2 controller so it could use the motherboard's dimm slots. AMD was talking about licensing coherent HT and about "accelerators" so this could be it...


K8L
By bob5820 on 3/18/2006 8:33:15 AM , Rating: 3
In an earlier article someone from AMD had stated that the up comming K8L CPU would have enhancements to its arithmatic functions. Obviously the possible Clearspeed deal is comming a little to late for the K8L, unless...
Could the Clearspeed silicon be integrated into the K8L already, and the anouncement has been timed for marketing reasons.




RE: K8L
By Furen on 3/19/2006 7:43:34 AM , Rating: 2
I doubt clearspeed silicon will be integrated because having to pay royalties with every chip made is certainly not worth it. Also, I'm pretty sure code has to be customized in order to take advantage of the clearspeed processors, so throwing the silicon on-chip doesn't make much sense. So I'd guess it'll either go into an HTX slot or into an Opteron socket. The Opteron-socket thing is what sounds most intriguing to me, since this means that if you buy a, for example, 8 socket system you can mix and match general purpose CPUs and FPU accelerators (and perhaps other types of accelerators later on). This sounds like the ultimate customizable system, and I'm sure the HPC crowd would just love it... regardless of what woodcrest brings to the table.


RE: K8L
By Viditor on 3/19/2006 11:29:51 AM , Rating: 2
I agree Furen...I'm thinking socket as well. At 8-32 sockets and 4 cores and 1 co-processor/socket, this sounds like an HPC dream machine design.


You guys don't get it yet...
By Viditor on 3/18/2006 9:33:40 AM , Rating: 2
This was mentioned months ago (vaguely) by AMD's CTO...
What they were talking about is giving a co-processor cHT links. A cHT link (or coherent HT) is the direct link to the AMD cache and memory. At 1400 MHz and direct access to memory and cache (no bus bottleneck), there isn't anything made that could touch it for scientific and heavy math processing.
This isn't going anywhere NEAR the desktop, this is for extreme high-end only. And Intel wouldn't even be in the same ballpark with these, this is a Power5 killer...




RE: You guys don't get it yet...
By stephenbrooks on 3/18/2006 10:03:37 AM , Rating: 2
25Gflops is a bit disappointing though - only 5x a general purpose CPU. I kind of thought vector processors would do dozens of times the performance (a bit like graphics cards do) because they're massively parallel.


RE: You guys don't get it yet...
By Knish on 3/18/2006 4:14:12 PM , Rating: 2
It looks like you can cram a few onto a PCB though.


Math coprocess
By ConsumerBoomer on 3/18/2006 12:46:17 PM , Rating: 2
Why can't they develop there own instead incorporating someone else's design?




RE: Math coprocess
By TomZ on 3/18/2006 3:32:33 PM , Rating: 2
Only AMD knows, but some reasons might include high development cost relative to small revenue potential for AMD and/or Clearspeed having proprietary technology (trade secrets and/or patents).


RE: Math coprocess
By Viditor on 3/18/2006 11:03:46 PM , Rating: 2
quote:
Why can't they develop there own instead incorporating someone else's design?

It limits their scalability. By providing cHT links to 3rd party copressors, AMD can enhance almost any specialized system tremendously!
For example, can you imagine how adding a dedicated JAVA coprocessor would enhance a SUN server? This way, AMD isn't limited to any single design...they just give the manufacturers of the coprocessor the specs and everybody (except Intel and IBM) wins!


Intel CSI
By abakshi on 3/18/2006 3:33:42 AM , Rating: 2
Wasn't CSI (Intel's HT-type bus) cancelled?




RE: Intel CSI
By beemercer on 3/18/2006 1:27:55 PM , Rating: 3
delayed until 2008 or 2009, something like that