Table 3 Speedup of GPU kernels
From: Effects of mesh loop modes on performance of unstructured finite volume GPU simulations
Mesh size | 1 million | 2 million | 4 million | 6 million | 9 million |
|---|---|---|---|---|---|
GPUGradientQ | 128.9 | 159.4 | 165.1 | 149.8 | 118.4 |
GPUNodeValue | 58.4 | 57.4 | 60.8 | 48.1 | 32.7 |
GPUViscousFlux | 468.5 | 443.1 | 455.1 | 393.4 | 295.1 |
GPUFluxSum | 141 | 142.1 | 161.3 | 141.7 | 108.7 |
GPUFaceValue | 350.2 | 346.9 | 320.5 | 280.1 | 218.7 |
GPULocalMinMax | 187.4 | 190.2 | 201.3 | 176.8 | 127.8 |
Average | 222.4 | 223.1 | 227.3 | 198.3 | 150.2 |