DailyTech - AMD Breaks Teraflop Computing Barrier

Submit News

Hardware AMD Breaks Teraflop Computing Barrier
Anh Tuan Huynh (Blog) - March 2, 2007 7:57 AM

30 comment(s) - last by crazydingo.. on Mar 3 at 10:39 AM

AMD demonstrates "Teraflop-in-a-Box" system powered by two "R600s"

AMD yesterday demonstrated a single system with over a teraflop of computing power at a press event in San Francisco, California. Next-generation AMD Stream Processors with a single Opteron dual-core processor powered the teraflop-capable system. The upcoming R600 graphics processor powers AMD’s next-generation Stream Processors. AMD’s “Teraflop-in-a-box” system ran Windows XP Professional.

“Today, teraflop computing capability is largely reserved for the supercomputing space. But now that “Teraflop-in-a-Box” is a reality, AMD can deliver an order of magnitude increase in performance,” said Dave Orton, executive vice president of visual media business, AMD.

AMD’s “Teraflop-in-a-box” system is capable of 1-trillion calculations-per-second when it comes to floating-point calculations. The capabilities of AMD’s “Teraflop-in-a-Box” marks a ten-fold performance increase over the 100-billion calculations-per-second capabilities of current high-performance servers. AMD expects scientific and commercial applications to benefit greatly from its next-generation teraflop capable R600 Stream Processors.

Release date and pricing information of AMD’s next-generation R600 Stream Processors are unknown and the demonstration is a proof-of-concept.

Comments

Threshold

Username
Password
remember me

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

see, this is what sony should have focused on

By scrapsma54 on 3/2/2007 8:45:13 AM , Rating: 1

cell means nothing if rsx cannot do that kind of calculations.
I think other gpu companies should learn from Ati's performance per watt gpu's. They maybe slower than nvidia cards but they can sure as hell pack a punch with lower wattage requirements.
Bah, ps3 is not going to show anything impressive, thats what they said about ps2.

RE: see, this is what sony should have focused on

By billythefisherman on 3/2/2007 9:01:51 AM , Rating: 4

Cell is a general purpose CPU much like Core Duo or Athlon (admittedly a little less conventional in design).

RSX is a GPU and so it's like comparing apples to oranges.

Parent

RE: see, this is what sony should have focused on

By scrapsma54 on 3/2/2007 9:11:32 AM , Rating: 3

I have seen many times over people saying ps3 is superior in graphics, yet the only thing superior about ps3 is its cpu. I know its 2 different things, yet people think more about cell than they do about rsx. Anyway, Amd did the right thing and striving for this kind of design, because performance per wattage is a necessity again.

Parent

RE: see, this is what sony should have focused on

By Griswold on 3/2/07, Rating: 0

RE: see, this is what sony should have focused on

By Scabies on 3/2/07, Rating: -1

RE: see, this is what sony should have focused on

By PrinceGaz on 3/2/2007 10:19:29 AM , Rating: 2

No it's not.

...a general purpose CPU. It has a single general purpose PPE, but is mainly composed of the eight limited function SPEs that are more like modern graphics-card shader units in functionality.

Parent

RE: see, this is what sony should have focused on

By ttnuagadam on 3/2/2007 10:24:43 AM , Rating: 2

nope, its not. theres ONE general purpose core in the cell (a very simple one at that) the other cores are purely for floating point performance.

Parent

RE: see, this is what sony should have focused on

By scrapsma54 on 3/2/2007 11:08:02 AM , Rating: 1

quote:
nope, its not. theres ONE general purpose core in the cell (a very simple one at that) the other cores are purely for floating point performance.

people can yell in our faces about it, but they arent going to change the fact that the cell does nothing for graphical quality.

Parent

RE: see, this is what sony should have focused on

By billythefisherman on 3/2/2007 4:41:39 PM , Rating: 2

Cell is the general purpose CPU in the PS3 it will help graphics by allowing more advanced physics, ai, graphical effects etc and help provide the vast quatity of data needed to feed the graphics chipset.

The PPE is just as complicated as any traditional CPU in its own right and the SPE's don't deal with just floating point numbers such as the VU's or VFPU seen in the PS2 and PSP respectively. They are however SIMD which enables them to process arrays of data extremely well but this doesn't stop them being general purpose.

http://research.scea.com/research/html/CellGDC05/i...

Watch out for my next game... ;)

Parent

RE: see, this is what sony should have focused on

By KernD on 3/2/2007 7:04:18 PM , Rating: 2

PPE is not as complicated as any of the single core processor you can buy for your PC, it's a small in order core, which means less efficient, just like the cores of the XBox360, you have to do more optimizing of your code by reordering instructions yourself and avoid branches.
On the 360 your often best to execute a useless instruction than to have an if to skip it, because the stall it creates is longer than executing the useless code.

And your right SPE don't just deal with floating points, they are more than that, but so are graphic cards processor, they can do some logic and branch instruction also...

The cell is nothing great or innovative, it's just a mix of two types of processor to get the best of both in one package, ho and by the way it has no place in the gaming world, the PS3 should have never had one, Sony just chose it so that they could continue there longstanding tradition of having the worst machine to code for of it's generation.

Parent

RE: see, this is what sony should have focused on

By billythefisherman on 3/3/2007 7:53:09 AM , Rating: 2

Yeah slighlty misleading in my previous post I was mearly pointing out the PPE provides a similar instruction set to that of modern desktop CPU where as a GPU such as the rsx does not - CUDA is great but I won't be trying to write a large app on it (yet): you think a general purpose 8 core Cell is hard to get the most out of...

Parent

RE: see, this is what sony should have focused on

By Scabies on 3/2/07, Rating: 0

RE: see, this is what sony should have focused on

By dqniel on 3/2/2007 10:55:33 AM , Rating: 2

Yeah, it's much to simple to be compared to a "conventional" CPU. It's not super-scalar from what I understand.

Parent

RE: see, this is what sony should have focused on

By defter on 3/2/2007 9:07:06 AM , Rating: 2

Where have you been? NVidia's G7x line has superior performance/watt compared to ATI's R5xx line. And according current information, R600 will consume significantly more power than G80.

Parent

RE: see, this is what sony should have focused on

By Stinger2 on 3/2/2007 11:49:42 AM , Rating: 2

Where is the rest of it?

Beyond3d was all over this yesterday.

Some more info:
"On the specifications front, several tidbits were pointed at to give enthusiasts some tea leaves to stir. In the configuration demoed in San Francisco, each R600 is said to have used 200W, be comprised of 320 "multiply-accumulate units", and between the two units on display be capable of a total of one teraflop of math power."

public override sealed string Sarcasm()
{
return "Yes, because 200W is significantly more than 180W.";
}

Parent

RE: see, this is what sony should have focused on

By Flunk on 3/2/2007 11:58:46 AM , Rating: 4

oooo, override and sealed you just got served Nvidia fanboy.

Parent

Wasn't this already done?

By jak3676 on 3/2/2007 8:55:44 AM , Rating: 2

I'll try to dig up the source, but I thought 2x G80's and any Kentfield already hit over 1 Teraflop. Most of the ops/second are coming from the video cards anyway, not the CPU's. (unless this is something totally different)

*OK, found it*
Not too sure on the source. But here we go. Are these numbers not right?
------------------------------------------------- -----------
The GeForce 7800 GTX have ~198 Gflops (July 2005);The Radeon X1900 XTX have ~375 Gflops (January 2006);The GeForce 8800GTX have ~520 Gflops (November 2006).

RE: Wasn't this already done?

By jak3676 on 3/2/2007 9:01:24 AM , Rating: 2

I forgot to say, I won't be real impressed until someone does 1 teraflop on a chip, not "in a box". My guess is that we'll be seeing that by this time next year. I think it will depend who refreshes their GPU archetecture next- AMD/ATI or Nvidia.

Parent

RE: Wasn't this already done?

By Griswold on 3/2/2007 9:47:47 AM , Rating: 1

You make it sound as if it matters if you're impressed with something you never need or utilize. This article is not about gaming performance, after all.

And no, it wasnt already done.

Parent

RE: Wasn't this already done?

By defter on 3/2/2007 9:09:40 AM , Rating: 3

Those numbers are right. G80 GTX has 128 stream processors running at 1350MHz, each processor can execute three floating point operations per second. 128*1350MHz*3 = 518.4GFlops

Parent

RE: Wasn't this already done?

By Griswold on 3/2/2007 9:58:48 AM , Rating: 2

Actually, I think it can only execute one MADD and one MUL operation per clock cycle. And thats what this article refers to. So, its not the 518.4GFlops that matters here in context of this article.

Parent

RE: Wasn't this already done?

By MrDiSante on 3/2/2007 8:05:47 PM , Rating: 2

It has been done. On a single chip mind you. http://www.dailytech.com/article.aspx?newsid=6057 - Intel 80-core experimental thing.

Parent

RE: Wasn't this already done?

By crazydingo on 3/3/2007 10:39:27 AM , Rating: 2

G80 is 346GFlops only.

As of now the MUL is not functional. If G80 was more than half a teraflop, then Nvidia would have already demonstrated a similar scenario.

With the way things are standing right now, either they need a super clocked G80 or a Quad-SLI G80 system (if they can shrink 8800 down to single slot)

Parent

Release it already!

By TejTrescent on 3/2/2007 8:53:07 AM , Rating: 2

Let the DX10 price war commence! I need to rebuild my system!

RE: Release it already!

By scrapsma54 on 3/2/2007 8:56:28 AM , Rating: 2

I bet by the time I get a dx10 computer the next windows will be out.

Parent

RE: Release it already!

By glennpratt on 3/2/2007 10:30:48 AM , Rating: 3

I laughed and rolled my eyes at this. Then I realized, I didn't get a DirectX 9 card until Vista RC2.

Parent

wtfsonybbq?

By WayneG on 3/2/2007 12:07:34 PM , Rating: 2

What the hell has this got to do with Sony in any shape or form?! The Cell for a start is an IBM CPU that is very very very basic in comparison to current Intel and AMD cpus, RSX is nvidia and is only a 7800GTX so how can you compare that to a next generation AMD CPU and a next generation graphics card...?! Cell and RSX are sh*t and the PowerPC CPU inside the 360 is also, however the Xenos (R500) is a thing of magic. Sony and or Microsoft have nothing to do with this article...

RE: wtfsonybbq?

By Scabies on 3/2/2007 12:22:20 PM , Rating: 3

in dailytech, everything has to do with the follies and failings of Sony, and the greediness and merciless nature of Microsoft. Duh.

Parent

Is this really a big deal?

By Jeff7181 on 3/2/2007 12:35:10 PM , Rating: 5

Didn't Microsoft already claim that Xbox 360 broke the "Teraflop in a box" barrier? And didn't everyone shoot them down because it was a theoretical number, never achieved in a real world test?

http://www.dvhardware.net/article4940.html

This begs the question...
Did AMD actually demonstrate this system's ability to reach 1+ Tflops... or did they demonstrate a system, which they claim can exceed 1 Tflop?

Few more updates

By subhajit on 3/3/2007 12:01:00 AM , Rating: 2

http://www.extremetech.com/article2/0,1697,2099613...
ATI is talking about 65nm mainstream parts.

"I'm an Internet expert too. It's all right to wire the industrial zone only, but there are many problems if other regions of the North are wired." -- North Korean Supreme Commander Kim Jong-il