System architects that live in the Seattle area who don’t want to uproot their lives and move to California or Texas or New York or maybe possibly Illinois or Oregon or even overseas to Japan or China have a fairly small number of job opportunities. But, the good news, as is illustrated by Steve Scott, the once and future chief technology officer at supercomputer maker Cray, is that they are really good opportunities.
Scott has just been named a Technical Fellow at Microsoft and corporate vice president of hardware architecture, which Scott confirmed to us in email. Microsoft has not made a formal announcement about this as yet, and he is not at liberty to say what the new gig entails, much as he was not able to say much about the little more than a year that Scott spent at Google as a principal engineer in the search engine giant’s Platform Group back in 2013 and 2014 after spending two years as chief technology officer of the Tesla datacenter GPU compute business at Nvidia from the summer of 2011 through the summer of 2013.
While Seattle has grown to become a center of cloud computing development in the past decade and a half, people forget that the stretch of geography in the old Northwest Territories in America – the one that included the area that would eventually include Wisconsin, Minnesota, and Illinois back in the late 1700s and early 1800s – is a hotbed for supercomputing. (Michigan and Ohio were also in this region but do not have the same supercomputing heritage.)
Supercomputing pioneer Control Data Corp, which was IBM’s main rival in the early days of supercomputing in the late 1950s, was located in Bloomington, Minnesota and the software part of that business still lives on this day as Ceridian. But significantly, Control Data, which started out selling clone magnetic drum disk drives (hence the name of the company) got into the mainframe business with none other that Seymour Cray as the chief architect of the CDC 1604 in 1958, following up with the CDC 6600, the world’s first supercomputer, in 1965. Cray, the man, set up a laboratory in Chippewa Falls, Wisconsin, the town where he was born, and it was there that the Cray Research company was formed and from there that the Cray-1 supercomputer came off the line in 1976, heralding a broader era in supercomputing and ending up at Los Alamos National Laboratory. Cray eventually split off another company, called Cray Computer, that was located in Colorado Springs, and portions of these two Crays ended up finally within Sun Microsystems (which was eaten by Oracle) and Silicon Graphics and Tera Computer. Tera, which was based in Seattle, renamed itself Cray after the acquisition, and both SGI and the final incarnation of Cray ended up being acquired by Hewlett Packard Enterprise over the past several years. HPE has a rich portfolio of supercomputing products and expertise at
Scott, who got is bachelors in computer engineering and PhD in computer architecture from the University of Wisconsin, was steeped in this supercomputing heritage and took his first job as a senior architecture engineer at Cray Research after graduating in 1992. When Cray Research was acquired by SGI in 1996, Scott stayed on for nearly four more years as a principal engineer and then chief engineer. After transputer pioneer Tera Computer bought the other Cray in 2000 and changed its name to Cray, Scott was hired to be chief architect and was named chief technology officer four years later. During that time, Scott was instrumental in designing the “SeaStar” family of interconnects that debuted in the “Red Storm” XT3 line of supercomputers launched in 2004, and followed that up with the “SeaStar+” interconnect used in the “Hood” XT4 system that debuted two years later. Scott also spearheaded the creation of the “Gemini” interconnect that came out in the XE5 and XE6 systems in 2008 and 2010, respectively. Both SeaStar and Gemini were 3D torus interconnects, and with the “Aries” interconnect that Scott led the design for, the “Cascades” XC systems based on Aries employed a dragonfly topology.
We have talked to Scott about networks and compute many times over the decades, and have always learned things from these conversations. In the early years of The Next Platform, Scott connected all the dots for us on various supercomputer interconnects, current and future, including his own as well as Ethernet, InfiniBand, and others. We talked about the possibilities and challenges in supercomputing with a Cambrian explosion in compute three years ago, and two years ago we were thrilled to see Cray get back into interconnects after a hiatus after selling off the Gemini and Aries networks to Intel in 2012, and Scott walked us through the details he could at the time here and then revealed some more about that new “Slingshot” interconnect in an interview at SC18 there. Cray’s former CTO gave a deep dive on the “Rosetta” ASIC that implements an HPC-augmented variant of Ethernet last summer at the Hot Interconnects conference.
Scott does change jobs ever now and then, as a lot of chip designers and CTOs do. And we are, like you, curious about what Microsoft might be planning with Scott’s appointment at Microsoft. As far as we know, the appointment is within the Azure public cloud organization, which has taken a shining to supercomputing in recent years, as evidenced by its adoption of InfiniBand in its clusters several years ago and the acquisition of Cycle Computing three years ago. It is interesting to contemplate Microsoft deploying Slingshot networks in Azure, but we suspect that Microsoft has much larger ideas in mind even if it does switch from InfiniBand to Slingshot for the HPC instances on its cloud because it would provide an Ethernet protocol across its entire network. You can sure bet that HPE is interested in selling Slingshot far and wide, and we hope that it is not silly enough to sell that business off. But if HPE does sell it off, AMD is the logical place to sell it, since AMD does not have an interconnect business any longer. (AMD did buy SeaMicro back in 2012 for its microserver interconnect, but didn’t do much with it for complicated reasons.)
What we know is that Scott has been instrumental in changing Cray’s supercomputers so they look more like clouds, and maybe now he can help Microsoft’s cloud look more like a supercomputer.
The real question is whether or not Scott will eventually have the trifecta of working for Amazon Web Services, Google, and Microsoft, the three biggest clouds whose operations are all located in the Seattle area.
In any event, as soon as Scott can talk about what he is up to at Microsoft – if that ever is even possible – we will give you the scoop.