Nvidia unveiled Rubin HPC on Monday at the SC26 supercomputing conference, a GPU for scientific simulation that carries more than 50,000 CUDA cores, which the company calls a record for a single GPU. It sits inside the Vera Rubin platform and is expected to arrive in 2028.

Nvidia’s pitch rests on agents. Software that proposes materials, molecules and designs produces a flood of candidates, and every one needs a simulation to test it. Ian Buck, who runs hyperscale and HPC at Nvidia as a vice president, said faster simulation means faster discovery for those agents.

The chip also adds sparse data acceleration meant to keep its cores supplied with data, plus seventh-generation NVLink for GPU-to-GPU connections. The announcement, written by Dan Ernst on the Nvidia blog, includes no benchmarks, pricing or power figures. The record is Nvidia’s own claim.

A 2028 launch leaves teams planning simulation capacity today buying current hardware, with Rubin HPC as a roadmap line rather than a purchase.

Nvidia blog (blogs.nvidia.com), by Dan Ernst, published November 16, 2026.