NVIDIA unveiled new accelerated data processing pipelines and Artificial Intelligence microservices at the SC25 conference aimed at advancing chemistry and materials science for industries such as aerospace, energy and manufacturing. Demonstrations in the NVIDIA booth highlighted the Holoscan Artificial Intelligence sensor processing platform for sub-10-nanometer visualization and new NVIDIA NIM microservices for batched conformer search and batched molecular dynamics. Those microservices are part of NVIDIA ALCHEMI, a suite of toolkits for chemistry and materials discovery, and are already in early access with companies including ENEOS and Universal Display Corporation.
Brookhaven National Laboratory is using the Holoscan platform with the National Synchrotron Light Source II to process high-bandwidth streaming X-ray data from beamlines and to provide near-instant feedback during scans. The Holoscan-accelerated pipeline lets researchers identify regions of interest on the fly, observe property evolution during measurements and run imaging workflows more efficiently. Faster edge processing not only reduces turnaround time for experiments but also helps optimize the operating costs and user throughput of expensive instruments. Brookhaven researchers see potential to integrate Artificial Intelligence-assisted operation for imaging tasks and experiment controls to enable more autonomous workflows.
Corporate users are applying NVIDIA ALCHEMI microservices to scale computational screening and simulation. ENEOS used the conformer search and molecular dynamics microservices to prescreen candidates for immersion cooling liquids and catalysts, evaluating about 10 million liquid-immersion candidates and 100 million oxygen evolution reaction candidates in a few weeks, at least 10 times more than prior methods. Universal Display Corporation used the conformer search to evaluate billions of OLED candidates up to 10,000 times faster than traditional methods, then ran molecular dynamics simulations up to 10 times faster per run and distributed workloads across multiple GPUs to reduce simulation time from days to seconds. NVIDIA ALCHEMI joins over 150 NVIDIA CUDA-X libraries and frameworks supporting accelerated science and engineering workflows.
