NVIDIA Details BioIR Runtime: 2.90x Higher Boltz-2 Folding Throughput, 58.5K Residues per GPU-Hour
AllThingsApx · x · 2026-09-11
NVIDIA has published a technical deep dive on BioNeMo Inference Runtime (BioIR), a Python library that accelerates structure-prediction models on NVIDIA GPUs while keeping the standard PyTorch workflow.
Key facts:
- Boltz-2 folding throughput improved 2.90x, reaching 58.5K residues per GPU-hour on 8xH100
- Already powered the AlphaFold Database expansion: protein-complex structures across 4,777 proteomes (31 million candidate complexes), with 1.81 million released as high-confidence predictions
The framing: the bottleneck has shifted from folding a single protein to moving proteome-scale worklists through parsing, featurization, GPU inference, and output writing. BioIR is open-sourced on GitHub with precompiled CUBINs; it requires Python 3.12+, a compatible GPU and driver, and a staged model checkpoint.
More from Infra
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27
- AI Infra Meme: While Countries Need TW-Scale Power for Datacenters, People Argue About Lightbulbs — mgostIH · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Cloud release velocity scores: AWS 95, Azure 90, Google Cloud 82 — DavidLinthicum · 2026-10-27
- Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
- AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27