NVIDIA: CUDA-X libraries cuPhoton, DAQIRI, and ALCHEMI accelerate astronomy, chemistry, and materials science

NVIDIA launches CUDA-X libraries for AI in science

NVIDIA announced three new software libraries from the CUDA-X ecosystem — a platform that accelerates GPU applications for specific scientific domains. The libraries cuPhoton, DAQIRI, and ALCHEMI target astronomy, detector physics, and chemistry/materials respectively, delivering dramatic speedups over classical CPU pipelines.

What does a 14,900× speedup mean in practice?

cuPhoton is a library designed for optical and radio telescopes: on an NVIDIA GB200 system it achieves 14,900× faster astronomical data loading and 8,400× faster signal processing for the LSST telescope (Legacy Survey of Space and Time). By comparison, the classical CPU approach that previously took hours now executes in seconds — a difference that changes the pace of astronomical research. cuPhoton and the VASP microservice are coming in summer 2026, while DAQIRI for real-time networking of physics detectors is already available on GitHub.

Results and availability

ALCHEMI — a suite of microservices for chemistry and materials — was integrated by Lila Sciences, achieving a 50× speedup in materials screening and 30% faster computation of magnetic properties. ALCHEMI is already available on GitHub and PyPI. More broadly, the NSF NAIRR program (National AI Research Resource) has supported more than 700 research projects over two years, providing a minimum of four DGX nodes per research team. The combination of CUDA-X libraries and NAIRR infrastructure marks a shift toward the democratization of GPU-accelerated science.