Steve Scott has given up his job as the chief technology officer for NVIDIA’s GPU business to help build big systems for Google. It’s unclear exactly what Scott, who formerly designed interconnects for Cray, will be doing at the Internet giant, but the smart money is that it’s something big.
Scott joined NVIDIA in 2011 as the company was ramping up its efforts to build Tesla GPUs and get them into the HPC market. With his deep level of understanding of HPC architectures, Scott was key in the company’s efforts to work with supercomputer makers like Cray and integrate the GPUs into their system architectures.
NVIDIA marked a different direction for the career of Scott, who spent 19 years designing supercomputer interconnects at Cray from 1992 to 2011. Cray sold its interconnect business, including the Gemini and Aries interconnects that Scott helped design, to Intel in 2012.
Now Scott’s career is taking him in another direction, although exactly which direction is not clear. When asked by Timothy Prickett Morgan, the systems editor at The Register, what he will be up to at Google, he reportedly replied that he would be “working on new Google systems” for the Silicon Valley firm, and left it at that.
But that leaves lots of room for speculation. Google runs some of the largest clusters on the planet, although they’re not of the supercomputer variety. The company designs and builds its own X86-based systems, and while performance is obviously critical to Google’s various businesses, the workloads are of a different sort than what Scott has worked on during the bulk of his career up to this point.
The speculation is that, in the future, as the technologies and techniques that HPC system makers have employed to bolster performance and throughput on big HPC workloads will start to bleed over into the general systems market, that people of Scott’s caliber will be needed to smooth the wrinkles that inevitably develop. Google is seen to be on the cutting edge of this trend, hence the desire to snag one of the brightest minds in the business.
Scott was in charge of the architecture and the roadmap for the Tesla GPU. NVIDIA will be relying on others to pick up the slack left by Scott’s departure, including CTO Bill Dally, GPU architect Jonah Alben, and Sumit Gupta, the general manager of the Tesla accelerated computing business unit at NVIDIA.