SIAgentNews.com
← All news

NVIDIA launches AI tools to speed up scientific research and data analysis

New GPU-accelerated libraries and open models aim to reduce complex research tasks, like black-hole data correlation, from months to minutes.

· First reported by interestingengineering.com

Blueprint drawing: picture for the story 'NVIDIA launches AI tools to speed up scientific research and data anal'
AI-made blueprint drawing for this story. No real people or products are shown.
900xspeedup for cell polygon calculations

The short version

  • The cuPhoton.xBeam toolkit reduces black-hole data correlation from over a month to just 8 minutes.
  • New Daviator GPU kernels can accelerate quantum-mechanical calculations by up to 22 times.
  • The cuSpa tool enabled a 900-fold speedup for cell polygon calculations at Memorial Sloan Kettering Cancer Center.
  • NVIDIA released Nemotron for Science, an open model package for chemistry reasoning and research tasks.
  • The new LitmusBench v0.2 dataset contains nearly 50,000 questions for training chemistry models.

Why it matters

These tools lower the barrier for complex scientific discovery by making high-powered simulations and data processing faster and cheaper. Researchers can now move from interpreting raw data to running complex experiments in real time.

Accelerating heavy data workloads

NVIDIA is expanding its CUDA-X libraries to provide specialized tools for scientific workloads that often face massive computing bottlenecks. These libraries are designed to help researchers handle the petabytes of data generated by large-scale instruments and simulations. By optimizing how hardware handles these tasks, the company aims to let scientists spend less time waiting for results and more time analyzing them.

One notable example of this is the cuPhoton toolkit, which focuses on processing astronomical observations. As interestingengineering.com reported, the cuPhoton.xBeam module allows researchers to correlate data from the Event Horizon Telescope in just eight minutes. Previously, this same correlation process took more than a month to complete.

The Event Horizon Telescope functions by linking radio dishes across the globe to create a virtual, Earth-sized observing instrument. Because the resulting data is so vast, it requires extensive processing power to turn into usable images. NVIDIA's new tools aim to streamline this pipeline, making it easier to study objects like black holes with greater efficiency.

Advancing drug discovery and materials

The company is also targeting the fields of drug discovery and materials science with tools that improve simulation accuracy and speed. For example, the cuPotential tool accelerates molecular simulation software by up to three times while improving the stability of calculations. This allows researchers to investigate how molecules bind and interact before they invest in expensive laboratory experiments.

In the realm of materials research, NVIDIA introduced Daviator GPU kernels for the Vienna Ab initio Simulation Package. These kernels can accelerate quantum-mechanical calculations by up to 22 times while simultaneously reducing costs by as much as 75%. These improvements are being integrated into the NVIDIA ALCHEMI platform to support broader scientific workflows.

Several companies, including Terray Therapeutics, Topos Bio, and Genesis Molecular AI, are already adopting these technologies. By providing faster ways to simulate molecular behavior, NVIDIA helps these organizations identify viable compounds more quickly. This shift helps move the needle from theoretical modeling to practical, real-world applications in medicine and engineering.

Open models for chemistry and biology

NVIDIA introduced Nemotron for Science, a package built around the Nemotron 3.5 Super VL model. This release includes a new open chemistry dataset and specific post-training recipes designed for scientific research. The goal is to help researchers build specialized agents that can reason through complex problems and use computational tools to perform specific tasks.

To support this, the company released LitmusBench v0.2, which contains nearly 50,000 questions for training and evaluating models on chemistry reasoning. This dataset draws on experimental drug-discovery data and molecular properties to help models predict how the human body processes different compounds. It specifically helps with predicting properties like solubility.

Organizations are already using these open models for diverse tasks. For instance, Terray Therapeutics used the model to extract chemical reactions from diagrams, while Therna applied it to design mRNA therapeutics. Argonne National Laboratory is also integrating the model into a multi-agent system to design and evaluate protein-based therapeutic candidates.

Real-time simulation and local control

A significant part of the new strategy involves allowing researchers to interact with simulations in real time. NVIDIA highlighted work with Pacific Fusion, where researchers used the cuPhoton.xLens module to extract insights while plasma simulations were running. When paired with a locally running Nemotron-powered agent, this workflow allows scientists to steer simulations as they happen.

To ensure data privacy and infrastructure control, NVIDIA optimized the Nemotron model to run locally on its 128GB DGX Spark system. This gives research organizations the option to work on hardware they manage themselves rather than relying on external clouds. This local execution is often critical for sensitive research or high-performance computing needs.

The combination of these computing tools and open models is intended to change how scientists work. Instead of just interpreting existing data, researchers can now run more complex computational experiments. By developing specialized scientific workflows, they can move faster from a hypothesis to a verified discovery in various fields.

What to watch

  • How other research institutions adopt the Nemotron model for specialized tasks.
  • The impact of the 22x speedup in quantum-mechanical calculations on materials science costs.
  • Future updates to the cuPhoton toolkit for other types of astronomical data.

Try it

Check out the Nemotron 3.5 Super VL model and the LitmusBench v0.2 dataset on Hugging Face to see the open-source components available for research.

Sources and links

First reported by interestingengineering.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news