Comments on: Ruminations About Europe’s “Alice Recoque” Exascale Supercomputer https://www.nextplatform.com/2024/06/25/ruminations-about-europes-alice-recoque-exascale-supercomputer/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Fri, 12 Jul 2024 16:00:32 +0000 hourly 1 https://wordpress.org/?v=6.5.5 By: Xpea https://www.nextplatform.com/2024/06/25/ruminations-about-europes-alice-recoque-exascale-supercomputer/#comment-226007 Wed, 26 Jun 2024 00:47:31 +0000 https://www.nextplatform.com/?p=144347#comment-226007 With Nvidia bringing a new compute system every year, I wonder how SiPearl can keep up and not being a liability instead of a benefit. Of course, it’s a matter of politics and technology independence but still, the effort must make sense. At this pace, with low financing, limited HPC experience and accumulated delays, SiPearl will only become one of these numerous companies kept alive by government subsidies, which is not a good situation to be in…

]]> By: Hubert https://www.nextplatform.com/2024/06/25/ruminations-about-europes-alice-recoque-exascale-supercomputer/#comment-226004 Tue, 25 Jun 2024 22:22:34 +0000 https://www.nextplatform.com/?p=144347#comment-226004 Great analysis! With Neoverse V2 Demeters already in the field (Nvidia Grace, AWS Graviton4, Google Axion), Rhea1’s Neoverse V1 does seem a bit behind the curve for a chip meant to be used in state-of-the-art supercomputers. Granted though that China’s exafloppers are likely built with chips that are also not at the bleeding edge, and that realizing tape-out and qualification of a first chip should be a great learning experience for improved future versions.

That being said, as much as the Fujitsu-Riken A64FX may be viewed as a Neoverse V0 (ushering in the Neoverse), developed tightly with ARM, I would hope that SiPearl, the Jules Verne Consortium (JV), and/or the EU Joint Undertaking (JU), could also work tightly with ARM to develop a Rhea2 or 3 chip that pushes Neoverse tech forward, beyond what ARM has already announced, to V4 for example. Otherwise, yes, as suggested in the article, just go for a CSS V3 Poseidon Voyager (ideal candidate) and be done with it promptly (especially for Alice Recoque’s Rhea2), before this fast moving tech passes by, and the JU/JV has to rely (again) on the usual suspects for its main CPUs.

]]>