Comments on: HPC Market Bigger Than Expected, And Growing Faster https://www.nextplatform.com/2024/11/19/hpc-market-bigger-than-expected-and-growing-faster/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Mon, 02 Dec 2024 21:04:35 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Slim Albert https://www.nextplatform.com/2024/11/19/hpc-market-bigger-than-expected-and-growing-faster/#comment-240294 Thu, 21 Nov 2024 06:07:08 +0000 https://www.nextplatform.com/?p=145020#comment-240294 Non-traditional suppliers (NTS) should be great sources for the innovative knives required at each TNP-classic trifurcation buffet of diverging performance requirements ( https://www.nextplatform.com/2022/10/06/the-art-of-system-design-as-hpc-and-ai-applications-diverge/ ) … lest one wants to go digital by computing on his/her actual fingers (best reserved for eating, in finger lickin’ good ways)! And at 21% CAGR for NTS, it’s a doubling every 4 years (a tad better than the “Note to self” 15% overall!).

I’m guessing Hyperion’s NTS include the outfits developing systolic dataflow near-memory machines as well as reconfigurable optical networking, that should help with AI and flexibly addressing workloads on the continuum from HPL to AI (along with MxP and HBM). But there’ll also be challenges with HPCG and graphs to tackle, for which innovations beyond MRDIMM might be needed (possibly retro-futuristic ones: https://www.nextplatform.com/2023/09/01/what-would-you-do-with-a-16-8-million-core-graph-processing-beast/ ).

It’s remarkable for example that under current trifurcated conundra, the CPU-only Fugaku is 6.8x faster than Frontier, and 8.5x faster than Aurora at Graph500, even as those two are, conversely, 3x faster than Fugaku on HPL ( https://graph500.org/?page_id=1332 ). Hopefully some NTS “knives” can make these diverse workloads operate with more consistent perf on future supermachines!

]]>