Histogramming on CUDA-LLM task suite
7.55Latency (ms)OptiML
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| OptiMLGPU=A100 80GB, Variant=G+X2026.02 | 7.55 | 1.07 | 34.38 | 41.37 | 25.44 | 917 | 3,012 | 3,392 | 76.19 | 52 | |
| HPC-Coder-V2 + OptiML-XGPU=A100 80GB2026.02 | 7.82 | 1.04 | 33.02 | 42.97 | 24.91 | 954 | 3,110 | 3,521 | 74.98 | 52 | |
| GPT 5.1 + OptiML-XGPU=A100 80GB2026.02 | 8.1 | — | 32.11 | 44.63 | 24.28 | 985 | 3,192 | 3,613 | 74.14 | 52 | |
| StarCoder2 + OptiML-XGPU=A100 80GB2026.02 | 8.34 | 0.97 | 31.21 | 44.88 | 23.74 | 1,006 | 3,265 | 3,659 | 73.06 | 52 | |
| Qwen2.5-Coder + OptiML-XGPU=A100 80GB2026.02 | 8.65 | 0.94 | 30.62 | 46.21 | 23.11 | 1,043 | 3,371 | 3,754 | 72.22 | 52 |