NVIDIA expands CUDA-X libraries and hands them to AI agents: Event Horizon Telescope correlation drops from over a month to eight minutes
Synopsis
NVIDIA has expanded its CUDA-X libraries with cuPhoton, cuPotential and cuSpa, and made them callable by AI agents to speed up scientific computing. cuPhoton.xBeam cuts Event Horizon Telescope correlation analysis from over a month to eight minutes; cuPotential accelerates existing molecular simulation software by up to 3x; the open-source cuSpa reaches a 900x speedup over CPUs for cell polygon calculations at Memorial Sloan Kettering. These figures belong to different libraries and different settings.
Interpretation
NVIDIA extends CUDA-X into astrophysics, fusion, molecular simulation, materials and biological data processing, and lets AI agents call these libraries to accelerate research workflows. These steps previously relied on CPUs or conventional pipelines; for example, Event Horizon Telescope correlation analysis took over a month, and data generated by fusion experiments often waited until the experiment ended for analysis. cuPhoton.xBeam shortens Event Horizon Telescope correlation analysis from over a month to eight minutes and successfully correlated observations of black hole jets J1924 and 3C279.
The libraries move heavy computation that once ran offline into GPU pipelines, so analysis can proceed while data is being produced. cuPhoton.xLens lets researchers at Pacific Fusion extract results while plasma simulations run, rather than after they finish. The entire pipeline, from data loading and processing to visualization, runs on distributed multi-GPU systems or a compact NVIDIA DGX Spark, and can be coupled with a locally running Nemotron agent for interactive analysis.
Acceleration in the molecular and materials directions is offered in a way that is compatible with existing software, rather than requiring researchers to switch toolchains. cuPotential builds on work by the Flatiron Institute and D. E. Shaw Research to accelerate existing simulation software. cuPotential accelerates existing simulation software by up to 3x while improving calculation stability and accuracy; Daviator provides batched GPU kernels for VASP, delivering quantum-mechanical calculations up to 22x faster and at up to 75% lower cost.
Perspective
The report covers the speedups and adopters NVIDIA announced for each library, including Event Horizon Telescope correlation, in-run analysis of Pacific Fusion simulations, molecular dynamics engine collaborations and cell polygon calculations at Memorial Sloan Kettering; each figure corresponds to its own library and setting.
The speedups come from NVIDIA, and the material does not specify the hardware configurations or baseline details used for measurement; Daviator's benefit for small systems depends on scale, and the material says such systems previously ran faster on CPUs.
