Acceleration of many small matrix calculations on GPU¶
Satoshi Ohshima (Kyushu Univ.)
Abstract¶
One of the important strategies of GPU computing is calculating many data in parallel. However, some of the numerical calculations don't have enough parallelism to fill the many computation cores of current GPU. In this talk, we show a case study of our matrix calculation. To fill the many computation cores and achieve high performance easy, we are focusing to use GPU by multiple processes.