Comments on: The Third Time Charm Of AMD’s Instinct GPU https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Wed, 16 Aug 2023 06:36:28 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Fyzhkar https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-212398 Wed, 16 Aug 2023 06:36:28 +0000 https://www.nextplatform.com/?p=142527#comment-212398 Is it just me or i think the memory configurations is either 16GB x 8 for the MI300A & 24GB x 8 for the MI300X, That’s why the X is sampling later compared to the A because 24GB HBM3 stack also came at a later date than the 16GB stack as per SK Hynix announcement (in which i believe the module is provided by them)

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210069 Sat, 17 Jun 2023 01:00:01 +0000 https://www.nextplatform.com/?p=142527#comment-210069 In reply to tufttugger.

I’ll take one.

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210068 Sat, 17 Jun 2023 00:59:42 +0000 https://www.nextplatform.com/?p=142527#comment-210068 In reply to tufttugger.

Yes. I am suggesting that the HPC flagship will overclock. The FP64 and FP32 vector were assuming virtual the same clock, which was my original assumption. Then I saw the official 850 watt rating and thought, maybe they are overclocking it….

]]> By: tufttugger https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210059 Fri, 16 Jun 2023 16:35:09 +0000 https://www.nextplatform.com/?p=142527#comment-210059 In reply to Timothy Prickett Morgan.

Somebody must be interested in the all CPU version for them to have that sku…

]]> By: tufttugger https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210058 Fri, 16 Jun 2023 16:33:46 +0000 https://www.nextplatform.com/?p=142527#comment-210058 In reply to Timothy Prickett Morgan.

Cool, thanks. But I don’t see a change/update yet in the chart. Is the result skewed partly due to the frequency you have the CUs running at? Thanks!

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210014 Thu, 15 Jun 2023 11:18:23 +0000 https://www.nextplatform.com/?p=142527#comment-210014 In reply to Dave.

I am looking forward to seeing how this all works. It would make one hell of a CPU with all Genoa cores….

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210013 Thu, 15 Jun 2023 11:17:45 +0000 https://www.nextplatform.com/?p=142527#comment-210013 In reply to Tufttugger.

A vestigial mistake from an earlier model I didn’t catch. The FP64 and FP32 are too low for this scenario. Fixed. And thanks.

]]> By: Dave https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210006 Thu, 15 Jun 2023 08:19:15 +0000 https://www.nextplatform.com/?p=142527#comment-210006 My favorite bit of the ingenuity of AMD is they have put the IO die into a base die with the infinity cache as a L3 cache below the GPU and CPU chiplets.

]]> By: EC https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-210000 Thu, 15 Jun 2023 05:39:42 +0000 https://www.nextplatform.com/?p=142527#comment-210000 The “Inference advantage” slide compares “number of GPUs” on the vertical axis. The blue line in the chart appears have under-counted by multiples of 8 for some odd reason.

]]> By: Tufttugger https://www.nextplatform.com/2023/06/14/the-third-time-charm-of-amds-instinct-gpu/#comment-209998 Thu, 15 Jun 2023 04:53:21 +0000 https://www.nextplatform.com/?p=142527#comment-209998 Why is the MI300A lower precision estimates higher than your MI300X estimates?

]]>