AMD to build hardware-agnostic programming after push from leading AI firms, says CEO Lisa Su

Liza Su, CEO and Chair, AMD

Chipmaker Advanced Micro Devices (AMD), on November 21, outlined the company’s strategy to create an open-source and hardware-agnostic software environment for artificial intelligence (AI) development.

"I think the desire for getting to hardware-agnostic programming environment is very high, and some of it comes from us as hardware vendors, then a lot of it comes from the users on the application standpoint. I do think that they're going to get more adoption because they're being driven by the leading AI companies out there," said Lisa Su, chief executive officer and chairman of AMD, at a fireside chat at the Indian Institute of Science (IISc), Bengaluru.

The Santa Clara-headquartered firm is seen as the world's largest chipmaker Nvidia's closest competitor for chips powering advanced data centers, which handle massive amounts of data for Generative AI (Gen AI) technologies.

Su, who is also the first cousin of Nvidia CEO Jensen Huang, said that the world needs an open-source software environment. "It shouldn’t matter whether it’s AMD or Nvidia or ABC as the hardware layer you want to build on top of that, and with the software under the abstraction.”

Su explained AMD’s significant investments in tools, compilers, and abstraction layers to support this vision, going forward. “We’re investing significantly in all of the tools and compilers and the abstraction layers that will allow you to build this open-source ecosystem, and I would say that there’s tremendous support in the industry for this type of more open ecosystem,” she added.

Discussing the importance of holistic system design in AI, Su said there is a need to balance performance, power, and cooling constraints. “When you talk about the environment that we have to build, it’s one where we consider all of the constraints, which include performance, power, and the overall cooling of these machines."

Also read: Why AMD chose India to set up its largest global design centre

The future of public-private partnerships is looking at these computing problems holistically, including hardware, software, overall systems environment, and keeping an opportunity for a lot of exchange between all of these groups.

The fact that the machine learning library PyTorch is rapidly growing is a testament to the demand for a hardware-agnostic framework, and its ability to support a large number of models across various platforms.

PyTorch is used in applications like computer vision and natural language processing, developed initially by Meta AI, and now operates under the Linux Foundation's umbrella. “The fact that PyTorch now runs more than a million models. You had asked me that question 12 months ago, it was probably a third of that number. So it’s ramping up very quickly,” Su said.

Moreover, she also pointed out that the largest foundation models, such as the newly released Llama models, have been running on day one across AMD, Nvidia, and other hardware platforms. This is because the industry is increasingly focused on making it easier to adopt and utilise diverse hardware solutions.

Also read: American chipmaker AMD eyes bigger role in India’s 6G journey