Summary
CUDA current augmented_cudss
689/1582
616 solved
| set | device | problems | Ipopt | current augmented_cudss | ||||
|---|---|---|---|---|---|---|---|---|
| solved | SGM (s) | SSGM | solved | SGM (s) | SSGM | |||
| ACOPF | CUDA | 132/132 | 123 | 19.57 | 1 | 126 | 5.51 | 0.525 |
| COPS | CUDA | 65/65 | 61 | 21.16 | 1 | 58 | 9.30 | 0.619 |
| CUTEst | CUDA | 445/1338 | 402 | 8.65 | 1 | 389 | 8.09 | 0.97 |
| Mittelmann | CUDA | 47/47 | 39 | 89.67 | 1 | 43 | 13.85 | 0.239 |
Solvers
| solver | device | revision | benchmark |
|---|---|---|---|
| Ipopt | CPU | 3.14.19 | – |
| current augmented_cudss | CUDA | 68f00afc | 0.1.0 |
CUDA: shin-compute-001: Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz (32 cores, 64 threads), 1006.5 GB of memory; Quadro GV100, 31.7 GB, driver 13.0.0, CUDA runtime 12.9.0; shin-compute-000: Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz (48 cores, 96 threads), 376.5 GB of memory; Quadro GV100, 31.7 GB, driver 13.0.0, CUDA runtime 12.9.0
Benchmark 0.1.0: the version fixes the instances, the options, Ipopt and HSL.
SGM: shifted geometric mean of the solve time, with a shift of 10 s and a failure counting as the time limit of a solve (CPU 7200 s, GPU 600 s, for every solver on those pages), over the problems that all solvers shown have run. SSGM: the SGM scaled so that Ipopt, or the best solver where Ipopt has not run, is 1.