Last week we announced the GPU-dense Cray CS-Storm cluster supercomputer, a highly innovative production system for application acceleration. This system is a milestone for Cray and will provide researchers and end users with one of the most high performance systems available for accelerated application workloads.
Let me describe the perfect storm of events that make NOW the right time for this type of solution.
Advanced Accelerator Technology
Accelerator technology is advancing rapidly. With its ultra-fast ECC on-package memory and full double-precision floating point performance, the NVIDIA® Tesla™ K40 GPU accelerator makes it the right technology for productive work.
Software Maturity and Innovation
Software ecosystems have matured to the point where productive and powerful applications and tools are now available.
A number of industries including energy, finance, machine learning and life sciences require advanced computing capabilities and specialized application processing. While there are solutions available that meet the needs of those requirements, none combine both advanced accelerator technology and today’s advanced software technologies.
Cray Design Expertise
Cray has been designing innovative technologies at the leading edge of HPC for the past 40 years, and the CS-Storm system is no exception. Key features include up to eight GPUs that work seamlessly with two-socket hosts, allowing for maximized bandwidth in and out of each GPU and advancing the performance of highly parallel accelerated applications. All this fits into the highest density 2U design available today and allows for power and cooling of the eight-GPU compute node with ease and finesse. In addition, this innovative design allows each on-board accelerator room to run at up to 300W and still maintain full power profiles. Although this is beyond what accelerators require today, we like to design for upgradability.
As you can see, we’re in the midst of a perfect storm and Grandma had better hold on to her seat!