The Pulse
NVIDIA expands CUDA-X tools for scientific AI agents
NVIDIA has expanded its CUDA-X scientific computing tools with software for astronomy, molecular simulations and spatial biology. The company says the tools can accelerate research workflows and, in some cases, work with AI agents.

Researchers working on astronomy, molecular simulations and biology have new and expanded GPU software tools in NVIDIA’s CUDA-X collection, including an open-source library for spatial biology and tools that can work with AI agents. NVIDIA described the additions in a company post on November 16, alongside claims of faster processing across several research workflows.
From telescopes to simulations
The Event Horizon Telescope links radio dishes around the world and produces petabytes of data. NVIDIA says its cuPhoton.xBeam module can correlate observations that previously took over a month to analyze in eight minutes.
A separate NVIDIA technical post reports that xBeam processed a 10 TB dataset in 70 seconds, about 460 times faster than an x86 CPU baseline of 31,992 seconds. That comparison used an NVIDIA Vera Rubin NVL72 system with 64 GPUs and GPUDirect Storage in compatibility mode. The technical post notes that constructing a 3D gas model from the resulting measurements is a separate downstream process that cuPhoton does not currently incorporate.
Pacific Fusion researchers have used another cuPhoton module, xLens, to extract insights while plasma simulations run. The pipeline can be coupled with a locally running NVIDIA Nemotron-powered agent for interactive analysis that may help steer simulations and inform fusion reactor target and chamber design.
Molecules and materials
For molecular simulations, cuPotential builds on work by researchers at the Simons Foundation’s Flatiron Institute and D. E. Shaw Research. NVIDIA says cuPotential accelerates existing simulation software by up to three times while improving calculation stability and accuracy. The company is working with Schrödinger and the AMBER, NAMD, GROMACS, LAMMPS and Open Molecular Software Foundation communities to bring it to molecular dynamics workflows.
Cadence, Fathom Therapeutics, Topos Bio, Genesis Molecular AI and Terray Therapeutics are adopting cuPotential for drug discovery. The software can also support synthetic data creation for generative AI models, identification of molecular binding sites and calculations of molecular interaction strengths, according to NVIDIA.
Daviator, a collection of batched GPU kernels for the Vienna Ab initio Simulation Package, is coming to NVIDIA ALCHEMI. NVIDIA claims the software delivers quantum-mechanical calculations up to 22 times faster and at up to 75% lower cost. At Lawrence Berkeley National Laboratory, the DAQIRI networking library connects an electron microscopy detector to GPU processing so researchers can analyze measurements during an experiment.
Tools for biological data
The newly open-source cuSpa library accelerates calculations involving cell polygons. NVIDIA says the tool enables developers to generate spatial graphs for patient and cancer models at Memorial Sloan Kettering Cancer Center, with a 900-fold speedup compared with CPUs. Its Python package assigns transcript coordinates to cell polygons and builds cell-by-gene count matrices on the GPU.
The rapids-singlecell framework, developed with scverse for single-cell analysis, is now available in a single agent-ready container. NVIDIA says Therna Biosciences processed data 200 times faster with it for therapeutic design, while 10x Genomics has integrated the framework into its Sentira platform.
An open-source GPU-accelerated HMMER library supports annotation of predicted protein sequences for biological foundation-model data preparation. The company also reports that benchmark searches at Basecamp Research ran up to 84 times faster than with standard HMMER.