Comments on: Mixed Results For The Datacenter Thundering Thirteen In Q4 https://www.nextplatform.com/2024/04/08/mixed-results-for-the-datacenter-thundering-thirteen-in-q4/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Tue, 16 Apr 2024 20:34:07 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Timothy Prickett Morgan https://www.nextplatform.com/2024/04/08/mixed-results-for-the-datacenter-thundering-thirteen-in-q4/#comment-222945 Tue, 09 Apr 2024 13:17:46 +0000 https://www.nextplatform.com/?p=143929#comment-222945 In reply to Hubert.

Nvidia has created the new System/360. There will be alternatives. Economics requires it. But provided we don’t have an extinction-level event, we think Nvidia will have a large and profitable share of AI compute and software for a long time. IBM’s System/360 got six decades and is still going. Nvidia will be strong as long as Jensen Huang is there. The minute he leaves, someone else will come in, just like the minute a Watson stopped running IBM then Big Blue started running up on the rocks in the late 1980s. It was just so big and profitable that it couldn’t feel it until maybe 1991.

]]> By: Hubert https://www.nextplatform.com/2024/04/08/mixed-results-for-the-datacenter-thundering-thirteen-in-q4/#comment-222937 Tue, 09 Apr 2024 09:27:32 +0000 https://www.nextplatform.com/?p=143929#comment-222937 Wow! The impact of Nvidia being (essentially) first to market with an (rather) integrated hardware+software solution for AI/ML is really well illustrated in this analysis (excellent!). Their focus on making low-precision compute very fast (in HW and SW) looks to have been a winning strategy in this space so far (relative to making FP64 fast), especially as datacenters became more curious about AI/ML capabilities and applications. The next few reports should tell us whether AMD is lagging by 2-3 quarters, or more, in this opportunity (IMHO).

It’s also fun to see the broad variety of HW and SW currently competing in this emerging AI/ML arena (eg. as reported by TNP). It is somewhat reminiscent of the CPU and OS diversity that existed before industry “standardized” mostly on x86 and Windows/Linux (also some POWER and Arm, and BSD, but not superscalar, superpipelined, Fairchild Intergraph Clipper …). It’ll be interesting to see which way AI/ML goes, either towards standardization, or sustained HuggingFace-like diversity, or one specialized arch and software for each type of AI app? Economics of scale would suggest standardization, and a majorly entertaining rhumbamageddon of doom between the major players!

]]>