Nvidia unveils Rubin HPC, a GPU aimed at simulation and data-heavy science work
Shipping is expected in 2028, and key specifications such as memory and FP64 allocation remain undisclosed.
Nvidia unveiled Rubin HPC at SC26, a new member of its Vera Rubin GPU family aimed at simulation and data-heavy analytics, which it says carries more than 50,000 CUDA cores on one GPU. The company pitches it as a balanced design that pairs compute with data movement, using seventh-generation NVLink, so that the chip's cores stay fed rather than idle during recurring workloads such as sparse matrix operations and reductions. Nvidia positions faster agent-driven simulation as a way to shorten the loop from hypothesis to refinement in climate research, genomics and financial risk. Rubin HPC is expected to ship in 2028, and Nvidia has not disclosed its memory capacity, bandwidth or FP64 allocation. Separately, Nvidia reported its first MLPerf Endpoints submission, where Vera Rubin NVL72 delivered more than 10x the throughput per watt of GB300 NVL72 on a provisional agentic workload, and it swept all MLPerf Training v6.1 benchmarks.
Why it matters
The announcement signals that Nvidia is extending its Vera Rubin line beyond AI inference into scientific simulation and analytics, where workloads such as climate research, genomics and financial risk depend on fast iteration. Because Nvidia has not yet published memory capacity, bandwidth or double-precision allocation, the practical gains for those fields cannot yet be judged. Buyers and researchers will need those figures before they can weigh the chip against existing hardware.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.
Sources · 3
- NVIDIA Blog ↗ Primary source
- NVIDIA Blog ↗ Primary source
- Tom's Hardware ↗