Ian Buck Kicks Off NVIDIA’s Week at SC26 as Science Enters the Age of Agents
AI-assisted digest, checked against the original source before publishing. How we work
At SC26 in Chicago, NVIDIA's Ian Buck unveiled agent tooling for scientific workloads, a new science-focused model release, and fresh HPC hardware.
NVIDIA kicked off its week at the SC26 conference in Chicago on Monday, with Ian Buck, the company's vice president of hyperscale and high-performance computing, delivering a special address. Buck announced Nemotron for Science, the Rubin HPC GPU, an expanded Vera CPU platform for high-performance computing, and a new quantum computing collaboration with IBM and Fermi National Accelerator Laboratory. He framed the moment as a shift in the scientific stack: "The frontier of computing and the new scientific stack is going to be simulation, as always, but combined with AI and the future of quantum."
Nemotron for Science pairs the Nemotron 3.5 Super VL open model with a newly released open chemistry dataset and post-training recipes aimed at life sciences and chemistry. Buck said researchers can run it locally on a DGX Spark, a workstation or their own supercomputer, rather than depending on hosted endpoints. The release positions agent tooling directly on the systems where simulations already run.
The address leaned heavily on demonstrated agent behavior rather than roadmaps. Buck described an agent at Argonne National Laboratory built with the toolkit that took a single prompt, planned and tested hypotheses, selected which simulations to execute, and ultimately identified two new binding sites. Separately, an agent from KTH Royal Institute of Technology that repairs broken GROMACS setups on its own fixed 81% of test cases during a single overnight run on one DGX Spark.
Other examples spanned radio astronomy and code porting. At the SETI Institute's Allen Telescope Array, an agent accepted a one-line prompt to plot a methanol line and then retuned for Doppler shift without further instruction. NVIDIA's new Nsight AI Porting Agent moved a routine from the Weather Research and Forecasting model onto GPUs, achieving more than a 5x speedup using roughly 150 million tokens. Buck anchored these in NVIDIA's two-decade CUDA history — he noted his first public CUDA talk was an SC06 tutorial in Tampa, and that the company now offers more than 1,000 CUDA-X libraries — arguing agents have become "a game changer" in how researchers interface with supercomputers, devices and instruments.
What remains open is how far this generalizes. The showcased results come from individual agents on specific tasks, and the source gives no detail on when Rubin HPC GPUs or the expanded Vera CPU platform ship, or on how the IBM and Fermilab quantum collaboration is structured. Watch for whether the Nemotron for Science recipes and datasets move beyond the reported demos, and whether the porting agent's token-heavy approach becomes economical across larger codebases.
Key facts
- • Ian Buck announced three products plus an IBM and Fermilab quantum collaboration at SC26.
- • Nemotron for Science bundles Nemotron 3.5 Super VL with an open chemistry dataset.
- • A KTH GROMACS-repair agent fixed 81% of test cases overnight on one DGX Spark.
- • An Argonne agent used one prompt to identify two new binding sites.
- • NVIDIA's Nsight AI Porting Agent delivered over 5x speedup on a WRF routine.
Why it matters
NVIDIA is packaging agents as on-premises research infrastructure — models, datasets and porting tools that run on a DGX Spark or a lab's own supercomputer — rather than as hosted services, which changes how compute budgets and data governance get planned. Builders should expect agent evaluation to shift toward overnight, single-GPU task suites with pass rates and token counts as the reporting met
Spotted an error? Send a correction.
Get the day's AI news in one email
The Daily Epoch · 60-second read · unsubscribe anytime