An Approach to the Highest Efficiency of the HPCG Benchmark on the SX-ACE Supercomputer
Authors: Kazuhiko Komatsu (Tohoku University), Ryusuke Egawa (Tohoku University), Yoko Isobe (NEC Corporation), Ryusei Ogata (NEC Corporation), Hiroyuki Takizawa (Tohoku University), Hiroaki Kobayashi (Tohoku University)
Abstract: This paper reports how we have achieved the world's highest efficiency
on the HPCG benchmark by fully exploiting the high potential of the
To achieve efficient vector calculations and memory accesses, several
optimization techniques such as ELL sparse matrix data packing, the
hyperplane method for eliminating data dependencies, selective caching
for the on-chip memory, and problem size tuning have been employed.
Evaluation results show that SX-ACE achieves an 11.4% efficiency
in the case of 512 nodes, which is the highest efficiency among all of
the supercomputers in the latest HPCG ranking.
This is because these optimization techniques can effectively exploit
the high potential of memory access performance of SX-ACE, which has
a strong impact on the overall performance of HPCG.
Two-page extended abstract: pdf