Parallel GPU/FPGA/CPU Gathering Environment and Application for Hybrid Accelerated HPC

Taisuke Boku
Seminar

CT

Argonne National Laboratory
Building 203 Auditorium (BH141)
Host Franck Cappello
Event Sponsor Mathmatics and Computer Science Division Seminar

GPU-ready supercomputers have been leading the accelerated HPC solution so far. It is one of the cost-effective and power-effective ways for certain categories of applications, but there are many cases where GPU acceleration is not perfect due to lack of high parallelism or irregular parallel behavior.

Center for Computational Sciences, University of Tsukuba, is now planning to introduce a new supercomputer, PACS-X, as a next-generation hybrid accelerated computing platform which introduces FPGA as well as GPU for computation nodes to provide a solution to combine GPU and FPGA in a side-by-side manner to compensate the weaknesses with each other. We are currently developing a framework to gather them, including GPU-FPGA communication module, FPGA-FPGA direct communication available on OpenCL-level, a demonstrating application on astrophysics, etc.

In this talk, I will present our current work as well Center for Computational Sciences, University of Tsukuba’s plan to introduce the PACS-X system which will start its operation in April 2019.

Biography: Dr. Taisuke Boku received his Master and PhD degrees from the Department of Electrical Engineering at Keio University. After his career as assistant professor in the Department of Physics at Keio University, he joined the Center for Computational Sciences (former Center for Computational Physics) at the University of Tsukuba where he is currently Deputy Director, HPC Division Leader and System Manager of Supercomputing Resources. He has been working there more than 20 years on HPC system architecture, system software, and performance evaluation of various scientific applications. In these years, he has been playing the central role of system development on CP-PACS (ranked as number one in TOP500 in 1996), FIRST (hybrid cluster with gravity accelerator), PACS-CS (bandwidth-aware cluster) and HA-PACS (high-density GPU cluster) as the representative for supercomputers in Japan. Also, He is a member of system architecture working group of Post-K computer development. H e is a coauthor of the ACM Gordon Bell Prize in 2011.