NVIDIA Details BioNeMo Inference Runtime (BioIR): 2.90x Higher Boltz-2 Folding Throughput and 58.5K Residues per GPU-Hour on 8xH100

NVIDIA has introduced BioNeMo Inference Runtime, a Python library designed to accelerate biomolecular structure-prediction models while remaining compatible with PyTorch. In internal benchmarks using the Boltz-2 model on 8xH100 GPUs, the library achieved 58.5K residues per GPU-hour, representing a 2.90x increase in folding throughput. This tool aims to improve the efficiency of computational biology workflows by optimizing how these specific scientific models utilize NVIDIA hardware.