Nvidia has released and expanded CUDA-X libraries aimed at scientific workloads that previously took too long to analyze during an experiment. The tools cover radio astronomy, plasma simulation, molecular modeling, materials research and biological data, with several components designed to be callable by AI agents.
The cuPhoton.xBeam module processed Event Horizon Telescope observations in eight minutes, according to Nvidia, compared with more than a month for the earlier workflow. Another cuPhoton module can analyze plasma simulations as they run, allowing a locally hosted Nemotron agent to help researchers explore results and adjust fusion studies.
For drug research, cuPotential accelerates existing molecular-simulation software by as much as threefold. Daviator provides batched GPU kernels for atom-level calculations and is reported to run up to 22 times faster at up to 75% lower cost. In spatial biology, the newly open-source cuSpa speeds cell-polygon calculations by 900 times compared with CPUs, while a GPU version of HMMER accelerated some protein-sequence searches by up to 84 times.
These figures come from Nvidia and named adopters, and they cover different hardware and tasks rather than one standardized benchmark. The practical shift is that some pipelines can move analysis closer to data collection, letting scientists inspect results before an experiment or simulation has finished.