NVIDIA launches BioNeMo Inference Runtime: 2.9x throughput for structure prediction, 1.6-3x in production
AllThingsApx · x · 2026-09-11
NVIDIA Health released BioNeMo Inference Runtime (BioIR), accelerating biomolecular structure-prediction models on NVIDIA GPUs while preserving the PyTorch workflow.
Technical highlights:
- End-to-end processor covering parsing, tokenization, feature generation, GPU inference, and PDB/mmCIF output, plus direct PyTorch module integration
- Three optimization layers: kernel selection, module optimization with CUDA Graph capture, and pipeline scaling via Ray replicas
- Benchmark: on 8xH100 across 1,000 human dimer targets, BioIR-accelerated Boltz-2 delivered 58.5K folded residues per GPU-hour vs 20.2K for a torch-compiled open implementation — a 2.90x improvement
- Energy: folding one million comparable targets costs 11 MWh vs 35 MWh for the public implementation
Production feedback: Proxima reports 1.6–1.9x throughput gains on H100 (up to 3x on short inputs) across distillation and refolding pipelines, with a coding agent wiring it into Neo in one shot via the BioNeMo agent skill.
Related event: NVIDIA Open-Sources BioNeMo Inference Runtime in Public Beta(9 posts)→
More from Infra
- 8x RTX 5090 training run hits ~$11/b tokens amid ~50% GPU failure rates — jon_durbin · 2026-09-11
- NVIDIA open-sources BioNeMo Inference Runtime to speed up protein structure prediction — AllThingsApx · 2026-09-11
- DeepSeek's New Model: 4x Smaller KV Cache Than DSV4-Flash and More Stable Training — stochasticchasm · 2026-09-11
- Reflect Orbital readies first satellite to sell sunlight, unfolding a volleyball-court-sized mirror in orbit — kyliebytes · 2026-09-11
- LLM inference bottlenecks: weight loading gave way to KV reads as contexts grew — YouJiacheng · 2026-09-11
- After GPUs and memory, AI agents are now driving a CPU shortage — The Pragmatic Engineer · 2026-09-11