SC16, the biggest annual supercomputing conference, is now taking place in Salt Lake City, Utah and one of the coolest new products being launched at the show is the Cray XC50 Supercomputer which Cray claims is able to reach over 1 petaflop of performance in a single cabinet.
In order to reach the 1 petaflop of compute performance, each one of Cray’s XC50 cabinets contain three chassis, which include 16 compute blades each. Each compute blade is then split into four nodes, each of which contain a Xeon CPU and a Nvidia Tesla P100 GPU Accelerator. By combining the total performance of 192 compute nodes, Cray’s XC50 is able to reach 902 teraflops through the GPUs and 141 teraflops on the CPUs producing a total of 1.04 petaflops per cabinet.
Of course if you’re a potential Cray customer, chances are you’ll need more than 1 petaflop of performance and according to Cray, hundreds of these cabinets can be setup together allowing the complete system to exceed 500 petaflops of total compute performance. To put this into perspective, currently the world’s fastest supercomputer, the Sunway TaihuLight at the National Supercomputing Center in Wuxi, China, is only capable of up to 93 petaflops.
Anybody buying one or more of these? Let us know in the comments below!