CUDA: use MMQ instead of cuBLAS by default (#8075) #13218
Job | Run time |
---|---|
3m 29s | |
9m 33s | |
3m 4s | |
1m 38s | |
2m 45s | |
3m 50s | |
4m 52s | |
1m 48s | |
2m 1s | |
8m 28s | |
5m 51s | |
1m 29s | |
2m 1s | |
3m 27s | |
6m 21s | |
8m 25s | |
13m 43s | |
11m 10s | |
5m 15s | |
11m 19s | |
7m 7s | |
2m 15s | |
2m 23s | |
9m 11s | |
1m 17s | |
8m 29s | |
7m 21s | |
33m 53s | |
18m 22s | |
6m 41s | |
13m 2s | |
34m 4s | |
8m 33s | |
12m 4s | |
7m 39s | |
18m 12s | |
7m 14s | |
6m 58s | |
7m 14s | |
2m 5s | |
3m 30s | |
1m 37s | |
5h 29m 40s |