Abstract:

Graphics Processing Units (GPUs) have emerged as a key technology whose
importance is likely to continue to increase in high performance
computing. Recent advancements in programmability embodied as CUDA and
OpenCL have made porting of existing CPU programs to GPUs possible in a
relatively simple manner. The TSUBAME supercomputer at Tokyo Institute
of Technology, currently ranked at the 56th in the TOP500 supercomputer
list, is one of the largest supercomputers in the world that exploit the
significant performance advantage of GPU. This talk will present the
overview of the current TSUBAME supercomputer and discuss our ongoing
research projects that are motivated with the deployment and operations
of TSUBAME. I will also present our future plans for the next generation 
TSUBAME that will further extend the scale of GPU acceleration while
incorporating various innovative technologies in power efficiency,
virtualization, and I/O performance.