NVIDIA cuPotential speeds up molecular dynamics 2-4x on existing GPUs without hardware switch
gdefabritiis · x · 2026-11-17
NVIDIA released cuPotential, a library of modular C API primitives that accelerates molecular dynamics by optimizing long-range interaction evaluation under periodic boundary conditions.
- Integrates with OpenMM via a plugin that drop-in replaces the nonbonded force, Langevin integrator, and Monte Carlo barostat
- Benchmarks on GPUs from T4 to RTX PRO 6000 Blackwell show end-to-end 2x to 3x+ speedups for systems of 24,000 to 400,000 atoms under NPT conditions
- New C2 continuous splitting/spreading functions developed with D. E. Shaw Research and the Flatiron Institute hit target accuracy at lower grid resolutions and cutoffs than standard Ewald
- Energy drift over 10 ns reduced by several hundred-fold vs default PME; dispersion energy accuracy improved several thousand-fold
Veteran developer Gianni De Fabritiis (ACEMD author, OpenMM contributor) called the 2x-4x speedup over OpenMM without a hardware switch unprecedented.
More from Infra
- Fact-checking the NVDA FUD: unopened Indonesia racks, $758B 2028 capex still intact — basedjensen · 2026-11-17
- Semantic Cache Pitfalls: Should LLM Caching Drop Dense Embeddings Entirely? — Artyrred · 2026-11-17
- A 10-item beginner reading list for LLM distributed training, from Horovod to FSDP — East-Muffin-6472 · 2026-11-17
- 18MB browser-run TTS model tiny-voca makes paid English TTS look silly — thisdudelikesAI · 2026-11-17
- Replicate quietly switches nano banana outputs from JPG downloads to presigned URLs — Kyrannio · 2026-11-17
- Fireworks AI tells investors chips are a commodity, its moat is software that runs models fast — hargup13 · 2026-11-17