NVIDIA announced Rubin HPC, a GPU for simulation and data processing, at SC26 on Nov. 16 and said it expects the chip to be available in 2028, according to a company blog post1.

The post gives no FP64 performance or memory bandwidth figures for Rubin HPC. Without them, the headline core count does not show how much simulation throughput the chip would deliver. The post also includes no independent test results.

Where Rubin HPC fits

Rubin HPC is a new GPU within NVIDIA's Vera Rubin platform. NVIDIA's platform page2 describes that platform as built for agentic AI and reasoning, and says it is "now in full production." That status applies to the platform's current systems, not to Rubin HPC.

Those systems use sixth-generation NVLink, which NVIDIA's NVLink page3 lists at 3 TB/s of bandwidth per GPU in preliminary specifications. Rubin HPC uses seventh-generation NVLink, according to the blog post, and is not expected until 2028.

NVIDIA already markets a Vera Rubin system for scientific work. The platform page says Vera Rubin NVL4, which connects four Rubin GPUs to two Vera CPUs, delivers up to 4x performance for scientific simulations2 compared with Hopper. That is a vendor claim about a different product, and the Rubin HPC post offers no comparable performance figure.

What NVIDIA says the chip does

NVIDIA says Rubin HPC has more than 50,000 CUDA cores on a single GPU, which it describes as a record. The company also cites a sparse data acceleration capability that it says keeps those cores supplied with data.

According to the post, the design scales data movement together with compute rather than increasing compute alone. NVIDIA says this accelerates recurring simulation patterns, including sparse matrix operations, stencil computations, sorting and reductions.

NVIDIA frames the chip around agents that test candidate materials, molecules and designs through simulation. "For agents tackling science and engineering, faster simulation means faster discovery," said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA.