Analyzing observations from the Event Horizon Telescope now takes just eight minutes, a process that previously required over a month, thanks to NVIDIA’s cuPhoton.xBeam module, the company says. This orders-of-magnitude speedup is enabling scientists to reconstruct a Messier 87* black hole galaxy in real time, having successfully correlated observations of black hole jets J1924 and 3C279. The expanded NVIDIA CUDA-X libraries, including cuPhoton, are accelerating research across multiple fields, giving AI agents the tools to help researchers make discoveries faster than ever.
cuPhoton Accelerates Black Hole Data Analysis & Astronomical Imaging
The module, cuPhoton.xBeam, within the larger cuPhoton Toolkit, dramatically accelerates the processing of data from the Event Horizon Telescope, a network that links radio dishes globally to create an Earth-sized aperture. This enhanced resolution allows scientists to study the behavior of matter under extreme gravitational forces, specifically how gas spirals inward toward a black hole. The entire processing pipeline, from initial data loading through visualization, now functions on distributed multi-GPU systems or a compact NVIDIA DGX Spark system.
The speedup achieved with cuPhoton is substantial; analysis previously requiring over a month now completes in eight minutes. This acceleration stems from the toolkit’s ability to combine data from the globally distributed radio dishes, effectively creating a telescope with resolving power. According to NVIDIA documentation, “By combining global radio dishes into Earth-sized aperture, cuPhoton delivers astronomical imaging resolution.”
The implications extend beyond simply faster processing. Researchers can now interactively analyze data and potentially steer simulations in real time to refine designs for fusion reactors. Coupling the data pipeline with an NVIDIA Nemotron-powered agent allows for this interactive analysis, informing the design of targets and chambers for fusion energy research, according to the company.
Beyond black hole imaging, the NVIDIA cuPhoton.xLens module provides real-time insights during plasma simulations, furthering understanding of the complex processes occurring within these extreme environments. This capability represents an advancement in the field, enabling a more dynamic and responsive approach to astrophysical research and potentially accelerating the development of sustainable energy sources.
cuPotential & cuSpa Speed Molecular and Cellular Research
Molecular simulations, essential for pre-clinical drug evaluation, are now significantly faster thanks to NVIDIA cuPotential. Built upon collaborative work with researchers at the Simons Foundation’s Flatiron Institute and D. E. Shaw Research, cuPotential accelerates existing simulation software by up to three times while simultaneously enhancing both calculation stability and accuracy. This improvement expedites the creation of synthetic data used to train generative AI models, allowing for the identification of previously unknown molecular binding sites and precise calculations of molecular interaction strength.
NVIDIA is actively integrating cuPotential acceleration into molecular dynamics engines and workflows in partnership with Schrödinger, AMBER, NAMD, GROMACS, LAMMPS and the Open Molecular Software Foundation communities. Several companies are already adopting cuPotential to advance their drug discovery efforts, including Cadence, Fathom Therapeutics, Topos Bio, Genesis Molecular AI and Terray Therapeutics.
The rapids-singlecell framework, codeveloped with scverse, further enhances single-cell analysis and is now available as a single, agent-ready container, the company says. 10x Genomics has integrated rapids-singlecell into its new Sentira platform, aiming to expedite the translation of large-scale spatial data into biological insights for academic researchers, biopharmaceutical teams and foundation model builders.
The open-source NVIDIA GPU-accelerated HMMER library is accelerating the annotation of predicted protein sequences, supporting data preparation for biological foundation models; at Basecamp Research, benchmark searches using HMMER ran up to 84 times faster than with standard methods, reducing the time needed to annotate the large biological datasets used to train therapeutic design models., the company states.




See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
