Table 6 Execution time(s) of GPU kernels with different mesh loop modes on K80
From: Effects of mesh loop modes on performance of unstructured finite volume GPU simulations
Kernels | Mesh 1 | Mesh 2 | Mesh 3 | Mesh 4 | Mesh 5 |
|---|---|---|---|---|---|
NV-F | 7.076 | 17.257 | 33.440 | 50.384 | 98.201 |
NV-C | 2.252 | 5.273 | 10.150 | 17.339 | 32.403 |
NV-N | 1.818 | 4.386 | 8.056 | 12.588 | 20.729 |
GQ-F | 0.603 | 1.362 | 2.503 | 3.974 | 7.694 |
GQ-C | 0.820 | 1.935 | 3.689 | 7.111 | 12.197 |
GQ-N | 6.496 | 15.970 | 30.262 | 45.027 | 79.358 |
SF-F | 0.840 | 1.825 | 3.443 | 5.703 | 9.890 |
SF-C | 0.618 | 1.440 | 2.656 | 4.400 | 7.580 |
LM-F | 0.338 | 0.751 | 1.371 | 2.241 | 4.553 |
LM-C | 0.154 | 0.354 | 0.655 | 1.089 | 1.980 |