Comments on: AMD Previews “Turin” Epyc CPUs, Expands Instinct GPU Roadmap https://www.nextplatform.com/2024/06/03/amd-previews-turin-epyc-cpus-expands-instinct-gpu-roadmap/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Thu, 13 Jun 2024 16:46:00 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: emerth https://www.nextplatform.com/2024/06/03/amd-previews-turin-epyc-cpus-expands-instinct-gpu-roadmap/#comment-225197 Tue, 04 Jun 2024 20:23:04 +0000 https://www.nextplatform.com/?p=144235#comment-225197 Pretty neat stuff. Some competition in the stratosphere. Now, will we ever see CDNA come back to us mortals in the small ISV space? AWS will rent me a VM with nV HPUs but not Mi300. Azure and Oracle claim to offer Mi300 but darned if I can find a way to provision such a VM. AMD should sell PCIe (a) cards with CDNA3/4 chips. It does not have to be a 1300 TF16 monster, 650 is fine. AI bubble is going to pop one day and when that happens AMD risks facing it’s old conundrum: all the software is written for CUDA. Few in the ISV world can afford the price of an 8 way MI300 machine… we need a reprise of the Radeon VII.

]]> By: Michael Bruzzone https://www.nextplatform.com/2024/06/03/amd-previews-turin-epyc-cpus-expands-instinct-gpu-roadmap/#comment-225195 Tue, 04 Jun 2024 19:56:26 +0000 https://www.nextplatform.com/?p=144235#comment-225195 Server and workstation component Market Share in time period that is December 2023 through May 2024.

AMD B+G+S+TR7 = 46.62%
Intel ER+SR+Atoms+RaptorE = 54.38%

AMD adds Milan+TR5 = 29.93%
Intel adds Ice+RocketE = 70.07%

AMD adds Rome+TR3 = 14.72%
Intel adds Cascades+Coffee Refresh E = 85.28%

AMD adds Naples+TR2+TR1 7.78%
Intel adds Skylake+CoffeeE = 92.22%

AMD aforementioned = 3.26%
Intel adds BW+HW+Ivy = 96.73%

To calculate cores share, to see current server SKU supply, trade-in and sales trend across 817 individual Epyc, Threadripper and Xeon server and workstation SKUs, to compare AMD and Intel directly competitive SKUs that takes a bit more work on channel supply data dropped into your own Excell file see Server Today and lift the data from the first table down;

https://seekingalpha.com/instablog/5030701-mike-bruzzone/5509267-server-today

On AMD Instinct accelerators, very few as in count the number on your fingers escape their commercial use base. Under AMD license provision? While Mi125 is readably available from open market sources Mi2x0 and M13x0_ are not. Certainly not as many as Nvidia Blackwell, Hopper and Ampere. If AMD accelerators were more readily available in the broker market and usually begins with academic use subject learning, perhaps AMD accelerators would be better understood. Then again that may not be what Instinct end customers want on their own application development tools and keeping uses proprietary to themselves. mb

]]> By: Mickey https://www.nextplatform.com/2024/06/03/amd-previews-turin-epyc-cpus-expands-instinct-gpu-roadmap/#comment-225153 Mon, 03 Jun 2024 18:30:11 +0000 https://www.nextplatform.com/?p=144235#comment-225153 “Catching up to and keeping pace with Nvidia in server GPUs is another thing entirely – and AMD has definitely done that and will be keeping pace for years to come”

How about asking for benchmarks between MI300 and H100 to start with? and, then make all the claims for future?

]]>