EU Grabs ARM for First ExaFLOP Supercomputer • Hacker News Board

Re: EU Grabs ARM for First ExaFLOP Supercomputer

#111

Post by janwas » Sun, Oct 08, 2023, 4:32 PM UTC

dotnet00 wrote: post #106

janwas wrote: post #104

Earlier quoted context omitted.

Golly. There are probably already enough comments in other threads saying this, but I'm curious how you would justify the description 'footnote'.

Footnote relative to the overall FLOPs. IIRC, on Perlmutter's GPU partition, 60 of its 64PFLOPs are represented by the GPUs, with the remaing 4PFLOPs, being the CPUs. In comparison, their previous system Cori, had ~3PFLOPs in the Haswell partition and ~30PFLOPs on the KNL partition. That, to me, indicates that CPU performance is mostly a footnote when GPUs are involved, as anyone who had previously been using Cori an…

OK, fair :) Do you have any data on how many tasks are actually compute-bound?

From where I sit, we're often limited by memory bandwidth. When a CPU such as A64FX or even M2 shows up with decent bandwidth, lo and behold, they are often competitive. I do not understand why we didn't see something like SPR Max years ago.

Re: EU Grabs ARM for First ExaFLOP Supercomputer

#112

Post by fsckboy » Sun, Oct 08, 2023, 7:56 PM UTC

monocasa wrote: post #39

owlbite wrote: post #8

It seems odd to focus on the CPU here, when presumably the vast majority of those flops re coming from the NVIDIA parts? Are the CPUs expected to contribute significant compute, as opposed to marshaling data in/out of the real compute units?

You'd be surprised. A lot of supercomputers aren't that much about individual CPU core perf, but having a lot of low power cores connected in a novel way. The BlueGene supercomputers were composed of low spec PowerPC cores (even for the time). High perf/watt matter more than just high perf/node, but even that balanced against 'how low latency can the interconnect be'. You then hit the high FLOP count with tons of nod…

I can believe what you are saying, but what are you saying, is it heat (ie cooling), power (cost), or hw cost/flop (because these chips are cheaper than screamers) that makes this an optimal solution?

Re: EU Grabs ARM for First ExaFLOP Supercomputer

#113

Post by yvdriess » Mon, Oct 09, 2023, 1:43 PM UTC

chasil wrote: post #42

ThinkBeat wrote: post #31

I get the impression this is more subsidizing domestic development than a decision made for the overall "best" CPU. https://www.eenewseurope.com/en/sipearl-raises-e90m-for-rhea...

However, the fact that Fujitsu previously claimed the top-performing "Fugaku" supercomputer with their own custom 48-core CPU (fabbed at TSMC) certainly justifies the choice of an AArch64 design. https://en.wikipedia.org/wiki/Fujitsu_A64FX

Sure, but they are still going for the (French) homegrown AArch64 solution instead of the Fujitsu design.

Re: EU Grabs ARM for First ExaFLOP Supercomputer

#114

Post by yvdriess » Mon, Oct 09, 2023, 1:48 PM UTC

orbifold wrote: post #60

eunos wrote: post #16

Earlier quoted context omitted.

Yea, much closer to Aachen in NRW. Side note, it's a small town without campus, I wonder why dont they locate the center in Aachen or Koln.

They used to do nuclear physics research (and still do to some extend) there and had an experimental fast breeder reactor (capable of producing weapons grade plutonium) on campus. They also prepared for rapid development of nuclear weapons capabilities in the 60s. It was a site for a potential German nuclear weapons program, I think they would have been able to produce enough material in ~6 weeks. They frame it as "n…

> You don't want something like that in a city centre.

Nuclear sites also have the tendency to be built on a nation's border.

Re: EU Grabs ARM for First ExaFLOP Supercomputer

#115

Post by monocasa » Mon, Oct 09, 2023, 8:29 PM UTC

fsckboy wrote: post #112

monocasa wrote: post #39

Earlier quoted context omitted.

You'd be surprised. A lot of supercomputers aren't that much about individual CPU core perf, but having a lot of low power cores connected in a novel way. The BlueGene supercomputers were composed of low spec PowerPC cores (even for the time). High perf/watt matter more than just high perf/node, but even that balanced against 'how low latency can the interconnect be'. You then hit the high FLOP count with tons of nod…

I can believe what you are saying, but what are you saying, is it heat (ie cooling), power (cost), or hw cost/flop (because these chips are cheaper than screamers) that makes this an optimal solution?

I'd describe the end goal as TCO of achievable FLOP.

So a lot of factors come out of that, and a lot of designs that take interesting stabs at new balances towards that goal.