Hybrid MPI and CUDA Parallelization for CFD Applications on Multi-GPU HPC Clusters

Table 1

Main performance parameters of GPUs.

	GTX 1070	Tesla V100
Date introduced	June 2016	January 2018
Architecture	Pascal	Volta
Computation capability	6.1	7.0
Device memory (GB)	8	32
Streaming multiprocessors	15	80
Stream processors	1,920	5,120
Single precision (TFLOP/S)	6	14
Double precision (TFLOP/S)	0.2	7
Memory bandwidth (GB/s)	256	900