When you think of Cray, chances are you are thinking about the company’s (somewhat overdesigned) machines from the ’70s and ’80s (most likely the circular ones with the benches around them). After a few ups and down and ownership changes in the ’90s, Cray has started to find its footing again, and its newest systems, like the XC and CS series, are now standards-based supercomputers that use Nvidia GPUs and Intel processors (and, on select models, this also includes FPGAs) to achieve peak performance of a petaflop or more in a single cabinet.
Unsurprisingly, those machines are still extremely expensive, and Cray’s focus remains squarely on the high-performance computing needs of researchers in both academia and the industry. These days, this also often means running machine learning jobs on these Cray machines.
You won’t be able to rent a Cray for just a few minutes to run your latest analytics batch job, though. Microsoft and Cray are planning to offer dedicated Cray systems in Microsoft’s data centers so that Cray’s users can easily access the rest of the Azure cloud services. Similarly, potential Cray users can now also get access to these machines without having to own and maintain a data center themselves. As a Microsoft spokesperson told me, each Cray system “will be custom configured to match the individual customer’s needs.”
“Using the enterprise-proven power of Microsoft Azure, customers are running their most strategic workloads in our cloud,” said Jason Zander, corporate vice president, Microsoft Azure, Microsoft Corp., in today’s announcement. “By working with Cray to provide dedicated supercomputers in Azure, we are offering customers uncompromising performance and scalability that enables a host of new previously unimaginable scenarios in the public cloud. More importantly, we’re moving customers into a new era by empowering them to use HPC and AI to drive breakthrough advances in science, engineering and health.”
It’s worth noting that this is the second time Cray has entered a partnership like this. The company is also making its machines available through Markley, a data center operator you may have never heard of, but that has been around since 1991 and currently manages more than 3 million square-feet of data center space across the U.S. and Europe.