From the source
NVIDIA announced Rubin HPC at SC26 on November 16, 2026, a simulation and data-processing GPU for science, engineering and analytics with more than 50,000 CUDA cores on one chip.
- The chip targets agent-driven workflows where agents generate candidate materials, molecules and designs that simulations then evaluate, a process NVIDIA says expands the simulation workload.
- NVIDIA identifies a shared pattern across climate modeling, engineering simulation, genomic analysis and financial risk: repeated calculations from small sets of values scattered through very large datasets.
- Rubin HPC gives data movement equal weight to arithmetic, with sparse-data acceleration meant to keep cores from idling, covering sparse-matrix and stencil calculations, sorting and reductions.
- CUDA-X supplies optimized routines agents can use as drop-in components, and seventh-generation NVLink extends the data-movement design to GPU-to-GPU communication across racks.
- Rubin HPC is part of the Vera Rubin platform, which NVIDIA says is ramping into full production with Vera Rubin-based systems shipping; Rubin HPC itself is expected to be available in 2028.