Metraipolis › archive › dc_6ffbf4c443b4
GPUPhysBench: Benchmarking Coding Agents for Correct and Efficient GPU Physics Simulation — arXiv:2609.35639
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35639
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:03:51.327Z
- SHA-256 of the archived text
4a73847c7b1d4e515c8382cf47397402db5fa71b6e2079d687563ed4d097eaa8- Stable URL of this page
https://metraipolis.com/documento/dc_6ffbf4c443b4
How to cite
GPUPhysBench: Benchmarking Coding Agents for Correct and Efficient GPU Physics Simulation — arXiv:2609.35639 — arXiv (en), https://arxiv.org/abs/2609.35639 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6ffbf4c443b4 (Metraipolis archive dc_6ffbf4c443b4, sha256 4a73847c7b1d4e515c8382cf47397402db5fa71b6e2079d687563ed4d097eaa8)
BibTeX
@misc{dc6ffbf4c443b4,
title = {GPUPhysBench: Benchmarking Coding Agents for Correct and Efficient GPU Physics Simulation — arXiv:2609.35639},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35639},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6ffbf4c443b4; sha256 4a73847c7b1d4e515c8382cf47397402db5fa71b6e2079d687563ed4d097eaa8},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35639. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Writing fast GPU code for physical simulation is difficult: implementations must preserve numerical accuracy while handling irregular data access, synchronization, and iterative solvers. We introduce GPUPhysBench, a benchmark of 50 tasks testing whether coding agents can meet these demands. Tasks cover fluids, deformable solids, and granular materials, from individual simulation operators to complete simulators. Agents write, compile, test, and optimize GPU code with access to a NVIDIA GPU under fixed time budgets. We report pass rates and runtime performance relative to expert-optimized reference implementations. In a single-attempt evaluation of six frontier model-harness pairs, the two strongest pass all 50 tasks, but even the fastest reaches at least 0.9 the reference speed on only 22% of them, and no submission is more than 5% faster than the reference. The largest gaps arise in collision detection, constraint solving, and iterative solvers. GPUPhysBench brings physical simulation workloads to coding-agent evaluation, testing both the ability to implement numerical methods correctly and the ability to make them run efficiently. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.35639 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T20:29:03.418Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).