Intel Aurora Supercomputer Breaks Exascale Barrier 28
Josh Norem reports via ExtremeTech: At the recent International supercomputing conference called ISC 2024, Intel's newest Aurora supercomputer installed at Argonne National Laboratory raised a few eyebrows by finally surpassing the exascale barrier. Before this, only AMD's Frontier system had been able to achieve this level of performance. Intel also achieved what it says is the world's best performance for AI at 10.61 "AI exaflops." Intel reported the news on its blog, stating Aurora was now officially the fastest supercomputer for AI in the world. It shares the distinction in collaboration with Argonne National Laboratory and Hewlett Packard Enterprise (HPE), which both built and houses the system in its current state, which Intel says was at 87% functionality for the recent tests. In the all-important Linpack (HPL) test, the Aurora computer hit 1.012 exaflops, meaning it has almost doubled the performance on tap since its initial "partial run" in late 2023, where it hit just 585.34 petaflops. The company then said it expected to cross the exascale barrier with Aurora eventually, and now it has.
Intel says for the ISC 2024 tests, Aurora was operating with 9,234 nodes. The company notes it ranked second overall in LINPACK, meaning it's still unable to dethrone AMD's Frontier system, which is also an HPE supercomputer. AMD's Frontier was the first supercomputer to break the exascale barrier in June 2022. Frontier sits at around 1.2 exaflops in Linpack, so Intel is knocking on its door but still has a way to go before it can topple it. However, Intel says Aurora came in first in the Linpack-mixed benchmark, reportedly highlighting its unparalleled AI performance. Intel's Aurora supercomputer uses the company's latest CPU and GPU hardware, with 21,248 Sapphire Rapids Xeon CPUs and 63,744 Ponte Vecchio GPUs. When it's fully operational later this year, Intel believes the system will eventually be capable of crossing the 2-exaflop barrier.
Intel says for the ISC 2024 tests, Aurora was operating with 9,234 nodes. The company notes it ranked second overall in LINPACK, meaning it's still unable to dethrone AMD's Frontier system, which is also an HPE supercomputer. AMD's Frontier was the first supercomputer to break the exascale barrier in June 2022. Frontier sits at around 1.2 exaflops in Linpack, so Intel is knocking on its door but still has a way to go before it can topple it. However, Intel says Aurora came in first in the Linpack-mixed benchmark, reportedly highlighting its unparalleled AI performance. Intel's Aurora supercomputer uses the company's latest CPU and GPU hardware, with 21,248 Sapphire Rapids Xeon CPUs and 63,744 Ponte Vecchio GPUs. When it's fully operational later this year, Intel believes the system will eventually be capable of crossing the 2-exaflop barrier.
Intel catches up to AMD (Score:2)
But with AI so they're totally better.
Intel not making any decent GPUs (Score:2)
So they have no other option than to stuff their HPC systems with nVidia cards, which happens to be the market leader when it comes to AI related cards.
AMD is stuffing their HPC systems with their own cards and they are still lagging behind nVidia.
Re: Intel not making any decent GPUs (Score:2)
Agreed, but at least they're not using other tech and claiming dominance. All Intel has these days is PR.
Re: Intel not making any decent GPUs (Score:2)
I hate that Intel is artificially limiting ram in their consumer cards to limit abilities to higher cards. AMD gives the ram but unfortunately that's because they're a generation behind. I'm not happy about it but I have a AMD CPU and a Nvidia GPU
Re: (Score:2)
I hate that Intel is artificially limiting ram in their consumer cards to limit abilities to higher cards. AMD gives the ram but unfortunately that's because they're a generation behind. I'm not happy about it but I have a AMD CPU and a Nvidia GPU
Is the complaint that low-end cards don't have high-end specs and features, or more so that high-end specs and features aren't available for low-end prices? Originally in the mid-2000's, the first guys figured out that video game GPUs could be used for GPGPU. Then the GPU companies decided to maximize revenue from deep-pocketed companies and government agencies. That's the obvious thing for the GPU companies to do, and it arguably is a good thing for the entire industry because it allowed those GPU compa
Re: (Score:2)
So they have no other option than to stuff their HPC systems with nVidia cards, which happens to be the market leader when it comes to AI related cards.
AMD is stuffing their HPC systems with their own cards and they are still lagging behind nVidia.
From what I've heard, Intel Arc cards actually aren't bad for AI stuff at their price point, They do better at compute-related tasks than games. They can't compete with the high end nVidia cards, though, as they aren't high end cards to begin with.
Re: Intel not making any decent GPUs (Score:2)
The shelf price is inconsequential compared to price/w which Intel doesn't excel at
Re: (Score:2)
Intel had plans to launch a DCG compute card called Ponte Vecchio showcasing their Intel 4 process as well as their advanced packaging technology. It did not go as planned. They scrapped plans to use their own internal nodes for most of the tiles in Ponte Vecchio, and chose TSMC to fab the silicon instead. Aurora was meant to showcase Ponte Vecchio (and everything I mentioned above). It has mostly failed. In any case, there was always more at stake than just building a fast computer.
Re: (Score:2)
Even if all you read was the summary, you can clearly see that they didn't catch up.
Aurora has been a comedy of errors.
Re:Intel catches up to AMD (Score:4, Funny)
Even if all you read was the summary, you can clearly see that they didn't catch up.
Aurora has been a comedy of errors.
Intel and Hewlett-Packard doing something together, what could go wrong...
Re: (Score:2)
^^^^ THIS ^^^^
How many doge coin flops is that (Score:2)
So Intel is inventing new definitions of flops? How about bitcoin flops ?
Re: (Score:2)
So Intel is inventing new definitions of flops? How about bitcoin flops ?
No, Nvidia, AMD, and everyone else are doing the same with different definitions of "AI flops" that are intentionally marketed with ambiguity.
How many small countries could this power? (Score:3)
Re:How many small countries could this power? (Score:4, Informative)
Frontier (AMD) 8,699,904 cores 22,786 kW
Aurora (Intel) 9,264,128 cores 38,698 kW
Re: How many small countries could this power? (Score:2)
Wait, so it's drawing 16MW more than the 2 year old AMD machine to not quite match its performance?
Re: (Score:1)
Wait, so it's drawing 16MW more than the 2 year old AMD machine to not quite match its performance?
Sounds about typical for intel since the P54c.
K6 was faster than a P2 when executing code optimized for it instead of pretending to to be an Intel chip, which admittedly was part of its job which it didn't do very well. Still, from K7 on there has seldom been any good reason to buy an Intel CPU.
Too little, too late (Score:5, Informative)
Per core performance 25% less than AMD's 2 year old machine. Per chip performance is even worse since AMD is 64 core and Intel is 52 core.
Even worse is almost twice as much power as AMD. Even the 2020 Fujitsu has a better power efficiency.
HPCG benchmark is even worse. AMD 2022 is almost 2.5x, and Fujitsu 2020 is almost 3x.
I can't see any dimension in which Intel is scoring better.
Re: (Score:2)
Actually the most interesting machine from Intel perspective is number 3 which is 560 Peta flops (about half of AMD) but it is using only quarter of the cores (coupled with NVIDIA H100). Even more interesting is that this was introduced last year with half the cores. They have doubled the cores but no change in performance (probably they didn't run it again). So about half the performance of the number 1 computer at 1/8th the cores.
Re: (Score:2)
Actually the most interesting machine from Intel perspective is number 3 which is 560 Peta flops (about half of AMD) but it is using only quarter of the cores (coupled with NVIDIA H100).
It's interesting because of the nvidia parts, not because of the intel parts.
Re: (Score:2)
It's using Sapphire Rapids, which was a very disappointing core. They did manage to match AMD's Genoa with their latest Emerald Rapids (well in per thread performance, they still don't have as many cores I think), which is good for competition.
Seen it, touched it. (Score:4, Interesting)
Units (Score:2, Funny)
Aurora supercomputer also secured the third spot on the high-performance conjugate gradient (HPCG) benchmark at 5,612 teraflops per second (TF/s) with 39% of the machine
Teraflops per second eh
Re:Units (Score:4, Funny)
Re:Units (Score:4, Informative)
Teraflops per second eh
1 Teraflop per second: 10 Teraflops after 10 seconds, 60 Teraflops after a minute.
This threw me off (Score:2)
However, Intel says Aurora came in first in the Linpack-mixed benchmark, reportedly highlighting its unparalleled AI performance.
And then there was me thinking that those machines are highly parallel.
Unfair (Score:2)
This supercomputer was given an illegal boost by the sun over the weekend.
Barrier? (Score:3)
This milestone isn't that. There's nothing different about a computer performing 0.99 exaflops versus 1.01 exaflops: the number is just a bit bigger. A rounding error could get you over the finish line and you wouldn't know the difference. Intel/Argonne didn't need to do anything substantially different to achieve an exaflop: they just threw more advanced cores at it (and a lot of money). Humans like round numbers, and inching over the next line tick mark on a graph is exciting, but that's really just an artifact of how we count things. 10^18 in some other number base would be...just a bunch of digits.