OpenAI CEO Sam Altman joined AMD chief Lisa Su on stage at AMD’s Advancing AI event, acknowledging that his company provided input as AMD was developing its Instinct MI400 Series GPUs, and is excited to see the results.
Whether Altman will have to wait for next year like everybody else remains to be seen, but he applauded AMD on building a GPU that is a whopping 10x greater AI performance that the company’s MI300X GPU and double the performance of the MI350X Series, along with 432 GB of HBM4 memory, an upgrade from the MI350 series’ HBM3E memory and in line with Nvidia’s anticipated Rubin GPU.
In what some company watchers described as a surprise appearance at the event, Altman said to Su on stage, “When you first started telling me what you were thinking about for the specs [for the MI400 Series, including the MI450], I was like, ‘There's no way. That just sounds totally crazy, but it's really been so exciting to see you all get close to delivering on this now. I think it's gonna be amazing.”
Altman also acknowledged that OpenAI had provided feedback during the GPU development process, saying, “We're already running some work on the 300X, but we are extremely excited [about the MI400 Series]. The work we've been able to do there as you've worked on it over the last couple of years, we're very grateful for the opportunity to have provided input. Hopefully, we have been a good representative for what the industry needs… The memory architecture is great for inference [and] for training as well.”
Jack Gold, president and principal analyst at J. Gold Associates, said it is not unusual for a big-name GPU user to provide such input. “It benefits both parties in that AMD can provide better products if they know what some of the issues are, and OpenAI gets products that are more optimized for their needs,” he told Fierce Electronics. “So it’s a win-win.”
The MI400X Series also will be at the heart of AMD’s new Helios system, which Su described as “a rack-scale, unified system” with 72 MI400X units and the company’s new Venice EPYC CPUs, connected via UALink, which could be seen as viable competition to what Nvidia is ding with its Blackwell rack systems.
Gold added, “My further take on AMD’s announcements is that they are making real progress on catching up to Nvidia, which should be much to the delight of major Nvidia customers. Having a sole source for products is very expensive and hard to manage. You have virtually no leverage in negotiations. If AMD becomes a true alternative, which in my opinion they already are for many needs, then price and delivery leverage returns to the customers.”