https://www.nextplatform.com/2024/01/10/sometimes-the-best-ai-accelerator-is-the-37888-gpus-you-already-have/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Mon, 22 Jan 2024 20:43:56 +0000 hourly 1 https://wordpress.org/?v=6.7.1 https://www.nextplatform.com/2024/01/10/sometimes-the-best-ai-accelerator-is-the-37888-gpus-you-already-have/#comment-218739 Thu, 11 Jan 2024 00:53:21 +0000 https://www.nextplatform.com/?p=143478#comment-218739 Why don’t they have any doors on their racks? I have a system like that where I work and we have doors…

]]> https://www.nextplatform.com/2024/01/10/sometimes-the-best-ai-accelerator-is-the-37888-gpus-you-already-have/#comment-218734 Wed, 10 Jan 2024 23:36:10 +0000 https://www.nextplatform.com/?p=143478#comment-218734 Did they say how long it took?

]]> https://www.nextplatform.com/2024/01/10/sometimes-the-best-ai-accelerator-is-the-37888-gpus-you-already-have/#comment-218723 Wed, 10 Jan 2024 19:26:04 +0000 https://www.nextplatform.com/?p=143478#comment-218723 Curiously we tested for a project beginning last year on Mi250X using AMD ACP cloud, and using PyTorch and DeepSpeed was completely transparent, with no code changes needed and huge speedup using the second.
But I’m sure doing on thousand of them it’s a complete different complexity.

]]>