AMD’s Helios strategy turns the GPU battle into a systems contest

The market for data center GPUs is evolving beyond individual chip specifications into a contest over fully integrated rack-scale systems. As AI workloads scale into the gigawatt range, buyers increasingly demand validated infrastructure that can be deployed quickly rather than components assembled piecemeal.

That shift has pushed Advanced Micro Devices Inc. to reposition itself as a systems company, unifying its data center GPUs, CPUs, networking and software into a single rack-scale platform called Helios. The company’s engineering roadmap and acquisitions, including ZT Systems, have compressed years of ecosystem development into a rapid systems buildout, according to Andrew Dieckmann (pictured), corporate vice president and general manager of the data center GPU business at AMD.

“I would agree with you; we’re a hardware company, a chip company at our core, but we have transformed into a systems company, and it’s really required in order to deliver the rack scale infrastructure that we’re doing with Helios,” Dieckmann said. “Whether it’s the design, the co-design with our customers, all of the design complexity of the rack itself, but also the operational needs to make sure that the entire supply chain is planned for the large production ramp of these systems and that all of the right components exist in the right places with the right quantities.”

Dieckmann spoke with theCUBE’s Dave Vellante and Bob O’Donnell at the AMD Advancing AI event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed AMD’s transformation into a systems company, rack-scale design tradeoffs and enterprise adoption of its data center GPUs. (* Disclosure below.)

Rack-scale design pushes data center GPUs past component specs

AMD builds modularity into its chiplet designs, rack architecture and open-source software so it can incorporate customer feedback despite compressed development cycles. The company favors deep collaboration with a handful of frontier model builders over broad simultaneous engagements, Dieckmann explained.

“We’re not trying to boil the ocean; we’re trying to be very selective in the market that we’re addressing and making sure that we’re getting the key elements delivered to deliver the best value to the customers,” he said.

On performance, AMD’s data center GPUs lean on memory capacity as a differentiator against competing chips. Because many workloads are memory-limited, extra high-bandwidth memory lets customers spread models across fewer GPUs, according to Dieckmann.

“We’ve tested this, we’ve retested it and we’ve tested it again,” he said. “Although the memory is more expensive now than it used to be, it is still of great value to our customers.”

Beyond hyperscalers, AMD is extending its reach through a new data center partnership with Cerebras Systems Inc., combining low-latency inference with Helios throughput to expand the addressable inferencing market, Dieckmann noted.

“It could be bigger. We’re watching that closely. There are some interesting use cases there,” he said. “I think these heterogeneous solutions like we announced today will be one way of expanding the total addressable market.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the AMD Advancing AI event:

(* Disclosure: TheCUBE is a paid media partner for the AMD Advancing AI event. Neither AMD, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE