From Materials Simulation to Experimental Astronomy, New NVIDIA AI Software Unlocks Scientific Discoveries

NVIDIA introduced a suite of software at the ISC conference in Hamburg designed to accelerate artificial intelligence workloads for scientific research, from materials discovery to astrophysics [1]. The new offerings, part of the NVIDIA CUDA-X collection, include the DAQIRI library, ALCHEMI NIM microservices, and the cuPhoton reference code [1]. The company states these tools convert workloads that previously required hours or days on CPUs into real-time, GPU-accelerated pipelines [1]. The field of artificial intelligence has seen massive advancements in generative models in recent years, with applications spanning industry and academia [2]. For astronomy, the cuPhoton code accelerates the processing of FITS data, the standard astronomical file format [1]. In early access testing, cuPhoton sped up the loading and reading of FITS images from the Rubin Observatory’s Legacy Survey of Space and Time by 14,900x [1]. It also enabled up to 8,400x faster signal processing and analysis using 32 NVIDIA Grace Blackwell superchips [1]. The Grace Blackwell superchip architecture shares its name with Grace Hopper, the American computer scientist and Navy rear admiral who pioneered machine-independent programming languages and developed the first compiler [6]. The DAQIRI library, short for Data Acquisition for Integrated Real-time Instruments, streams data from fast detectors into NVIDIA software to prevent data loss when instruments produce information faster than it can be saved [1]. A research project called A-GHOST, developed by scientists from CERN, the University of Chicago, and University College London, uses DAQIRI to run AI in real time on collision data from the ATLAS Experiment [1]. The system analyzes data that would normally be rejected due to storage constraints, which accounts for over 99% of the total [1]. In materials science, the ALCHEMI NIM microservices accelerate chemical and materials discovery for applications including battery materials, catalysts, and OLED displays [1]. Lila Sciences, a company building an autonomous lab platform, used the ALCHEMI NIM microservice for batched geometry relaxation to accelerate high-throughput materials screening by 50x [1]. The company then accelerated the calculation of magnetic properties by 30% for shortlisted candidates using an early-access ALCHEMI VASP microservice [1]. ALCHEMI’s specialized kernels for TensorNet provided Lila Sciences with a 6x speedup in training and inference and reduced memory usage by 3x [1]. Andy Beam, cofounder and chief technology officer of Lila Sciences, said the work “showcases using a powerful computing stack assembled to accelerate discovery at a scale no individual scientist could achieve alone” [1]. The DAQIRI library is available now on GitHub, while cuPhoton is expected later this summer [1].