From Materials Simulation to Experimental Astronomy, NVIDIA's New AI Software Accelerating Scientific Discovery

At the ISC conference, NVIDIA unveiled new software to accelerate scientific research in chemistry, materials discovery, dark matter search, and more. This includes the NVIDIA DAQIRI library, NVIDIA ALCHEMI NIM microservices, and the upcoming NVIDIA cuPhoton reference code. These tools reduce tasks that previously took hours or days on CPU-based systems to near real-time speeds through GPU-accelerated pipelines.

Key technologies and achievements include:

• NVIDIA cuPhoton: Accelerates the processing of astronomical data (FITS format). It improved data loading and reading speed for the Rubin Observatory's LSST data by 14,900x, and can boost signal processing and analysis speed by up to 8,400x when using 32 NVIDIA Grace Blackwell superchips.
• NVIDIA DAQIRI: A high-performance networking library that streams large volumes of sensor-generated data in real time without loss. CERN's A-GHOST project uses this to capture meaningful signals within the 99% of collision data from the ATLAS experiment that would otherwise be discarded.
• NVIDIA ALCHEMI: A toolkit for materials discovery, including batteries, catalysts, and OLEDs. The recently released BGR (Batched Geometry Relaxation) and BMD (Batched Molecular Dynamics) NIM microservices enable simultaneous simulation of millions of molecular structures and movements.