This article first appeared in Data Sheet, Coins2Day’s daily newsletter on the top tech news. Sign up here.
Advanced Micro Devices CEO Lisa Su wowed supercomputer geeks this summer when she showed off the company’s Project 47 at the SIGGRAPH conference in Los Angeles. In a rack just slightly taller than the MIT-trained Ph.D. CEO herself, Project 47 crammed 20 of AMD’s newest Epyc microprocessor chips inside along with 80 graphics processing units based on the company’s new Vega design and 10 terabytes of memory. The bottom line was a supercomputer that could perform 1 petaflop of calculations per second at a less-than-supercomputer price.
At a petaflop, which is a one followed by 15 zeros worth of calculations per second, that’s as fast as the fastest and priciest supercomputers in the world from 2009 and would rank just outside of the top 100 today. It’s also a testament to how fast progress has come in the past few years of chip designs, especially for graphics processors, which used to be relegated to the domain of video game junkies. And the biggest cloud data center companies, including Amazon (AMZN) and Microsoft (MSFT), are filling up more racks with GPUs as customers seek to run the kinds of artificial intelligence and machine learning tasks that thrive on the graphics processing architecture.
So far, its been Su’s fellow Taiwanese-born CEO, Nvidia’s Jensen Huang, who has captured the bulk of the new cloud GPU market. But AMD (AMD) just started shipping its newest Vega graphics chips in volume in the third quarter, the company revealed on Tuesday in its third quarter results. Along with Epyc, the server CPU that started shipping earlier this year, the chips are among the results of Su’s years-long effort to remake AMD’s entire line up with higher performing, modern designs. And though Nvidia (NVDA) has a big head start, and offers a popular and proprietary set of software tools to go with its chips, Su is optimistic that AMD can win a lot more business in the data center, too.
One advantage for Su may be that while Nvidia pitches GPUs and Intel (INTC) is selling CPUs, only AMD has leading products in both segments. Some cloud servers will use Epyc and Vega together, just like in the Project 47 supercomputer, she said.
“We see them separately and together of some systems,” Su told me this week. “It’s a fast growing segment of the market, and we’re very underrepresented in share, so we see that continuing to grow.”
