HPR Halpern Peaceman–Rachford
GPU-accelerated solvers for convex optimization problems
HPR-QP
SOLVERA GPU-accelerated solver for large-scale convex composite quadratic programming.
Download
Scroll
GPU-accelerated solvers for convex optimization problems
A GPU-accelerated solver for large-scale convex composite quadratic programming.
Page updated
65 instances · CPU and GPU results on their respective test platforms
HPR-LP-C (v0.1.4)
| Solver | Type | SGM10 (s) | Relative | Solved |
|---|---|---|---|---|
| HPR-LP-C | GPU | 10.8 | 1.35 | 63 / 65 |
| COPT-H* | GPU | 8.02 | 1.00 | 64 / 65 |
| cuOpt* | GPU | 10.9 | 1.36 | 62 / 65 |
| cuPDLPx | GPU | 28.5 | 3.55 | 57 / 65 |
| COPT | CPU | 25.3 | 3.15 | 65 / 65 |
| MOSEK | CPU | 89.2 | 11.12 | 56 / 65 |
| HiGHS | CPU | 256 | 31.92 | 55 / 65 |
| KNITRO | CPU | 372 | 46.38 | 48 / 65 |
| PDLP | CPU | 421 | 52.49 | 50 / 65 |
| XOPT | CPU | 146 | 18.20 | 59 / 65 |
* These configurations combine CPU and GPU computation or run multiple solver methods concurrently.
Open source: HPR-LP-C, HiGHS, PDLP, cuOpt, cuPDLPx · Commercial: COPT / COPT-H, MOSEK, KNITRO.
20 large-scale instances · Single-GPU comparison
HPR-LP-C (v0.1.4)
| Solver | SGM10 (s) | Relative | Solved |
|---|---|---|---|
| HPR-LP-C | 272.45 | 1.00 | 18 / 20 |
| cuOpt* | 1,376.58 | 5.05 | 13 / 20 |
| COPT-H* | 2,190.48 | 8.04 | 6 / 20 |
* These configurations combine CPU and GPU computation or run multiple solver methods concurrently.
Open source: HPR-LP-C, cuOpt · Commercial: COPT-H.
Source: H. Mittelmann, LPfeas Benchmark & Addendum. Published . Accessed .
