NVIDIA Debuts Accelerated Computing Libraries for Research at ISC

At the ISC conference underway in Hamburg this week, NVIDIA rolled out fresh software designed to boost AI performance across scientific fields that include chemistry, materials discovery and the hunt for dark matter.

New tools accelerate data pipelines from hours or days to real time. The NVIDIA DAQIRI library, newly released ALCHEMI NIM microservices and the forthcoming NVIDIA cuPhoton reference code reportedly transform lengthy CPU tasks into GPU-powered real-time workflows. All three belong to NVIDIA CUDA-X, described as a broad set of libraries and tools that deliver dramatically higher performance in AI as well as high-performance computing.

These improvements produce tangible results. Scientists across multiple disciplines now reportedly generate data and extract insights from instruments and large-scale surveys far more quickly than in the past.

cuPhoton speeds astronomical data processing by orders of magnitude. When deployed on NVIDIA GB200 NVL72 systems, cuPhoton accelerates the loading, reading, processing and analysis of FITS data — the standard file format used by observatories and telescopes. In early access testing it delivered a 14,900x speedup for loading and reading FITS images gathered during the Rubin Observatory’s Legacy Survey of Space and Time (LSST). The code also reportedly achieved up to an 8,400x improvement in signal processing and analysis when run on 32 NVIDIA Grace Blackwell superchips.

Faster pipelines allow quicker insights from the LSST camera, which is described as the largest digital camera ever built and records images of billions of distant galaxies plus faint nearby objects that reflect little light. NVIDIA cuPhoton functions as reference code that helps researchers pull insights from multidimensional data captured by telescopes, X-ray sources and laser experiments. It can load, process, analyze and visualize petabytes of information while combining with additional CUDA-X components to create complete accelerated pipelines for astrophysics and astronomy work.

Princeton and Harvard researchers will apply cuPhoton to dark energy surveys. Princeton University scientists worked with NVIDIA to create cuPhoton and plan to deploy it, together with Harvard University colleagues, to handle and examine the enormous datasets produced by observatories and dark energy surveys.

DAQIRI prevents data loss from high-speed instruments. Short for Data Acquisition for Integrated Real-time Instruments, NVIDIA DAQIRI serves as a high-performance networking library that streams information directly from rapid detectors and sensors into NVIDIA software stacks. Legacy hardware often loses data when instruments generate output faster than it can be stored. DAQIRI reportedly keeps pace by managing the incoming stream without delay.

Within the CERN openlab framework, the A-GHOST project — created by researchers from CERN, the University of Chicago and University College London — relies on DAQIRI. It performs real-time AI inference on collision data recorded by the ATLAS Experiment at CERN. The system examines more than 99 percent of the data that ATLAS would normally discard because of storage limits, thereby revealing potentially valuable signals that would otherwise go undetected.

ALCHEMI targets chemical and materials discovery. NVIDIA ALCHEMI consists of specialized microservices paired with a toolkit aimed at speeding chemical and materials research. Target applications span battery materials, catalysts, OLED displays, beauty products and additional domains. NVIDIA introduced two ALCHEMI NIM microservices in March for batched geometry relaxation (BGR) and batched molecular dynamics (BMD). The AI-driven utilities let teams simulate millions of molecules and materials in parallel so BGR can identify the most stable configurations while BMD models their motion across time.

ALCHEMI is also slated to add support for the widely adopted Vienna Ab initio Simulation Package (VASP) in the near term, which should raise GPU throughput for materials modeling.