NVIDIA opened its week at the SC26 supercomputing conference in Chicago with new tools for AI agents and scientific computing, including an open model for science, a new GPU built for research, and an expanded CPU platform.

Nemotron for Science pairs the open Nemotron 3.5 Super VL model with a new open chemistry dataset and post-training recipes for life sciences and chemistry. Researchers can run it locally on a DGX Spark, a workstation or their own supercomputer. At Argonne National Laboratory, an agent built with the toolkit planned and tested hypotheses from a single prompt and identified two new binding sites.

Other agents highlighted in the address include a GROMACS agent from KTH Royal Institute of Technology that repairs broken simulation setups on its own, fixing 81% of test cases in an overnight run on a single DGX Spark; a SETI Institute agent that plotted a methanol line at the Allen Telescope Array from a one-line prompt; and NVIDIA's Nsight AI Porting Agent, which moved a routine from the Weather Research and Forecasting model to GPUs and sped it up more than five times.

Rubin HPC is a new GPU in the Vera Rubin platform built for science, engineering, analytics and AI. It supports precisions from FP64 to FP4, adds sparse-data acceleration and connects over seventh-generation NVLink. With more than 50,000 CUDA cores, it has about three times as many as Hopper, and it is expected in 2028.

On CPUs, the German Climate Computing Center (DKRZ) has selected Vera CPUs for its HLRE-5 supercomputer, expected in early 2028, joining Vera systems already announced at the Texas Advanced Computing Center.

In quantum computing, NVIDIA, IBM and Fermilab announced a co-design platform pairing DGX B200 systems with IBM quantum computers, including IBM Quantum Nighthawk. A simulation on Julich's JUPITER supercomputer showed magic state cultivation can cut the operations needed for one logical step tenfold, and a new CUDA-Q decoding algorithm, TCBP-JOBS, cuts error-correction research computation from 12 weeks to 45 minutes. NVIDIA also showed one of the first deployments of its Photonics co-packaged optics switches, at Lambda, which it says are 15 times more reliable on bit errors and link flaps.