🔬NVIDIA Ships AI Agents That Run Research Experiments on Your Own Hardware
Your lab's simulations can now run themselves overnight
TL;DR
NVIDIA launched Nemotron for Science and new agent tools that plan experiments, run simulations, and check results on researchers' own machines. One agent repaired broken GROMACS setups on its own, fixing 81% of test cases in a single overnight run. If you run HPC workloads, this changes what a workstation can automate.
NVIDIA just gave research teams agents that plan experiments, run code, and check the results without someone babysitting them. The announcement bundles Nemotron for Science, the Rubin HPC GPU, an expanded Vera CPU platform, and a new quantum collaboration with IBM and Fermilab. The agent piece is the one to pay attention to. If you maintain simulation pipelines, this hits your workflow directly. An agent built with the toolkit took a single prompt, formed hypotheses, picked which simulations to run, and found two new binding sites at Argonne National Laboratory. The same toolkit repaired broken GROMACS setups by itself. Hot take: the boring glue work of keeping HPC jobs alive is exactly where agents will stick first. The numbers matter here. The GROMACS repair run hit 81% of test cases on a single DGX Spark overnight. A Nsight AI Porting Agent moved a WRF routine to GPUs for more than 5x speedup using about 150 million tokens. The Rubin HPC GPU carries roughly three times the CUDA cores of Hopper, with hardware slated for 2028. Watch the token budget, though. Those 150 million tokens are real money on any metered API.

Key Points
Nemotron for Science pairs the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes aimed at life sciences and chemistry workloads.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment