Comments on: AMD Is The Undisputed Datacenter GPU Performance Champ – For Now https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Mon, 24 Feb 2025 08:25:59 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Abel https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-249797 Mon, 24 Feb 2025 08:25:59 +0000 https://www.nextplatform.com/?p=143369#comment-249797 The evolution of AI accelerators is fascinating, especially with AMD’s latest advancements. Optimizing a high-performance data center is crucial as computational demands grow. I recently worked with a cloud solutions provider to enhance resource allocation and efficiency, which significantly improved system stability. For those exploring tailored branding solutions for cloud environments, here’s a helpful resource.

]]> By: Mickey Pearson https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-226349 Thu, 04 Jul 2024 20:04:04 +0000 https://www.nextplatform.com/?p=143369#comment-226349 So, it’s been what 7 months and there are no published benchmarks on performance or the increase in revenue would you Mr. Tim reconsider this? We understand the need for another horse in this race, but you seemed to claim the donkey a horse a bit tad too early no?

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-225934 Mon, 24 Jun 2024 00:58:12 +0000 https://www.nextplatform.com/?p=143369#comment-225934 In reply to Abhishek Joshi.

This story referred to performance, not revenue. So, no.

]]> By: Abhishek Joshi https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-225932 Sun, 23 Jun 2024 22:45:26 +0000 https://www.nextplatform.com/?p=143369#comment-225932 In light of AMD GPU sales for the past 6 months, would you reconsider this? Clearly, NVIDIA won and was the champ

]]> By: davide445 https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-218282 Sun, 31 Dec 2023 04:38:28 +0000 https://www.nextplatform.com/?p=143369#comment-218282 In reply to Mike Butash.

AMD acquired Pensando in 2022

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-217435 Mon, 11 Dec 2023 14:15:40 +0000 https://www.nextplatform.com/?p=143369#comment-217435 In reply to emerth.

You can’t and I ain’t selling mine. They really are the GPU of cars.

]]> By: Robert https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-217434 Mon, 11 Dec 2023 13:36:34 +0000 https://www.nextplatform.com/?p=143369#comment-217434 In reply to Timothy Prickett Morgan.

Isn’t that what GigaIO is doing, basically a PCIe based network switch?

]]> By: emerth https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-217428 Mon, 11 Dec 2023 09:04:23 +0000 https://www.nextplatform.com/?p=143369#comment-217428 Thanks Timothy! Great rundown of these chips. The 300A really looks cool for us mere mortals. Would love to see uniprocessor or dual socket options. Now, where can I get a Dodge Challenger for a decent price?

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-217333 Fri, 08 Dec 2023 14:11:15 +0000 https://www.nextplatform.com/?p=143369#comment-217333 In reply to JT.

Thank you Jim.

]]> By: Timothy Prickett Morgan https://www.nextplatform.com/2023/12/06/amd-is-the-undisputed-datacenter-gpu-performance-champ-for-now/#comment-217331 Fri, 08 Dec 2023 14:01:06 +0000 https://www.nextplatform.com/?p=143369#comment-217331 In reply to EC.

I think if it can run PyTorch and LLaMA 2, it is pretty tuned up and AMD can sell every one it can make at the same “price” adjusted for features as Nvidia is selling at.

And right now, that might be enough. There are people who believe that the game will not be more parameters, but a lot more data. For enterprises, this Meta Platforms stack might be enough, and become the preferred way on a few thousand GPUs running custom expert models for each business. They are not going to buy 25,000 or 50,000 GPUs and replicate what Google and AWS/Anthropic and Microsoft/OpenAI are doing with trillions of parameters and trillions of models.

Here’s the thing: A few tens of thousands of enterprise customers paying a profitable margin to AMD and its OEMs is a hell of a lot smarter than trying to sell to the Super 8 at a discount for their internal models. The clouds will love AMD+PyTorch+LLaMA for the same reason–more margin for them because enterprise customers will pay a premium compared to cloud builders and hyperscalers due to lower chip volumes bought or rented.

]]>