Parallel GPU/FPGA/CPU Gathering Environment and Application for Hybrid Accelerated HPC
Abstract: GPU-ready supercomputers have led accelerated high-performance computing (HPC) effort. They are one of the most cost- and power-effective ways for certain applications, but GPU acceleration is not perfect because of its lack of high parallelism and irregular parallel behavior.
The Center for Computational Sciences at the University of Tsukuba plans to introduce a new supercomputer, PACS-X, as a next-generation hybrid accelerated computing platform that introduces FPGA as well as GPU for computation nodes to provide a solution for combining GPU and FPGA in a side-by-side manner to compensate for their weaknesses. We are developing a framework to combine them, including a GPU-FPGA communication module, FPGA-FPGA direct communication available on OpenCL-level, and a demonstrating application on astrophysics.
In this talk, I will present our current work as well as the plan to introduce the PACS-X system, which will begin operation in April 2019.
Bio: Taisuke Boku received his Master and Ph.D. degrees from the Department of Electrical Engineering at Keio University. He is deputy director of the Center for Computational Sciences at the University of Tsukuba