News AMD-powered El Capitan is now the world's fastest supercomputer with 1.7 exaflops of performance — fastest Intel machine falls to third place on To...

[Admin]

AMD and the Lawrence Livermore National Laboratory announced today that the AMD-powered El Capitan has taken the top spot on the semi-annual Top500 list as the fastest-known supercomputer on the planet with 1.7 exaflops of performance.

AMD-powered El Capitan is now the world's fastest supercomputer with 1.7 exaflops of performance — fastest Intel machine falls to third place on To... : Read more

 

[Gururu]

Berkeley making the US proud!

 

[DS426]

Impressive that AMD has advanced so quickly in datacenter hardware that pure AMD CPU-GPU systems like this and Frontier are #1 and #2. Of course, the icing on the cake is LLNL stating that this was the most affordable way to get to these kinds of numbers. No Intel, no nVidia -- NOT surprised! Lol.

 

[gg83]

I love this stuff. Not too long ago we were talking about petaflops as the target.

 

[redgarl]

I don't understand how could Intel still maintain 75% of the market after projects like these.

 

[thestryker]

It definitely makes sense that the MI300A is delivering on both the design and performance front. For HPC installations the simplicity of a single chip has got to help a lot. It will be really interesting to see what AMD has next in this form.

 

[bit_user]

Before anyone gets too proud, just know that the AI bros have sailed way past these metrics. The only reason clusters like Musk's Colossus aren't leading the Top 500 is that they just don't bother to submit their systems to it.

 

[SBKch]

xAI cluster is easily 2-5 times faster than this, they just don't bother testing it at these type of workloads and adding it to top500 list.

 

[LaminarFlow]

Berkeley making the US proud!

LLNL, not LBNL

 

[SBKch]

xAI cluster is easily 2-5 times faster than this, they just don't bother testing it at these type of workloads and adding it to top500 list.

Also Collosus is 5-10 times faster than Aurora at AI workloads.

 

[einheriar]

I love this stuff. Not too long ago we were talking about petaflops as the target.

mwahhh we need to move to zetta flops, so we can focus onto reaching yotta flops..

 

[Gururu]

LLNL, not LBNL

Meh I went to school there, all the same.

 

[LaminarFlow]

Meh I went to school there, all the same.

And I work there, not the same 😛

 

[Gururu]

And I work there, not the same 😛

As a branch of the original Berkeley laboratory, I think we can still say it does Berkeley proud, as a child might make its father proud 😛

 

[Mama Changa]

70% more FLOPs than the Intel based sc and ~25% less power with ~20% more cores.

 

[abufrejoval]

Before anyone gets too proud, just know that the AI bros have sailed way past these metrics. The only reason clusters like Musk's Colossus aren't leading the Top 500 is that they just don't bother to submit their systems to it.

Same hold true for quite a lot of Chinese HPC systems, evidently: nobody wants to draw unwanted attention...

 

[jg.millirem]

Same hold true for quite a lot of Chinese HPC systems, evidently: nobody wants to draw unwanted attention...

AMD and the Lawrence Livermore National Laboratory announced today that the AMD-powered El Capitan has taken the top spot on the semi-annual Top500 list as the fastest-known supercomputer on the planet with 1.7 exaflops of performance.

AMD-powered El Capitan is now the world's fastest supercomputer with 1.7 exaflops of performance — fastest Intel machine falls to third place on To... : Read more

It’s a sad indictment of the technology world that many of these systems’ primary use is weapons, weapons, and more weapons. With a horde of tech careerists cheering it on and boasting.

 

[Ryrynz]

Fastest computer in the world, used to simulate effects of nuclear weapons.
AI in a decade or so: smartest intelligence in the known universe: Used to simulate and perform warfare.
AI in a couple of decades: Kills all Humans cos it just makes sense resource wise.

 

[bit_user]

It’s a sad indictment of the technology world that many of these systems’ primary use is weapons, weapons, and more weapons. With a horde of tech careerists cheering it on and boasting.

Not all supercomputers, even ones at the National Labs, are used only for weapons development & maintenance.

On this page about the Frontier supercomputer, scroll down to the section titled Scientific Impact. It lists 14 different profiles of scientists and their research (scrolls horizontally), many of which are non-military.

Frontier

www.olcf.ornl.gov

 

[JakeTheMate]

I don't understand how could Intel still maintain 75% of the market after projects like these.

Firstly. Inertia, takes awhile for the old systems to eventually be kicked off the bottom of the list.

Secondly, IIRC, sole source. When El Capitan, Frontier and Aurora were commissioned it couldn't be from a single supplier of the compute. Doesn't make sense to me when HPE built all three systems.

 

[JakeTheMate]

Before anyone gets too proud, just know that the AI bros have sailed way past these metrics. The only reason clusters like Musk's Colossus aren't leading the Top 500 is that they just don't bother to submit their systems to it.

Would they beat these systems on the high precision benchmarks as well or only the lower precision AI specific benchmarks? (if I understand the metrics correctly)

 

[waltc3]

I want one of these at home...😉 Bah, humbug...Not yet on Newegg...😉

 

[gg83]

mwahhh we need to move to zetta flops, so we can focus onto reaching yotta flops..

Very fast the machine will be.

 

[adamXpeter]

Before anyone gets too proud, just know that the AI bros have sailed way past these metrics. The only reason clusters like Musk's Colossus aren't leading the Top 500 is that they just don't bother to submit their systems to it.

From the article:

It's also important to note that supercomputer system performance in HPL is measured with full double-precision FP64. In contrast, AI-centric supercomputers are measured with smaller data types that enable much higher 'AI exaflop' ratings, but those aren't directly comparable to the listings on the Top500 list.

 

[adamXpeter]

Also Collosus is 5-10 times faster than Aurora at AI workloads.

Aww, really? Where are the test results shared?