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

Table 3

The runtime for GTX 1070 multi-GPU clusters.

No.	Two GPUs (ms)	Three GPUs (ms)	Four GPUs (ms)
Mesh 1	12.62	11.27	10.38
Mesh 2	20.86	18.19	16.01
Mesh 3	35.56	28.19	23.38
Mesh 4	67.55	52.43	42.33
Mesh 5	126.93	96.24	77.15
Mesh 6	252.01	187.84	144.43
Mesh 7	499.02	369.68	276.86
Mesh 8	—	—	540.73