ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
NVIDIA Unveils Rubin HPC: Over 50,000 Cores per Card, Aimed at Research Simulation

NVIDIA Unveils Rubin HPC: Over 50,000 Cores per Card, Aimed at Research Simulation

AI information • Admin • • 30 views

On November 16, 2026, NVIDIA announced Rubin HPC at the SC26 supercomputing conference in Chicago: a new GPU in the Vera Rubin platform with more than 50,000 CUDA cores on a single card, built specifically for simulation in science, engineering, and data analytics. Availability is expected in 2028.

Who the 50,000 cores serve: agents propose, simulation filters

Rubin HPC is positioned differently from the Rubin chip that trains large models: it targets a workflow where agents propose candidates — new materials, molecules, designs — and GPU simulation checks which ones are worth taking into the lab. Beyond the core count, NVIDIA gave it sparse data acceleration, so scattered data can keep up with what the cores consume, and seventh-generation NVLink for GPU-to-GPU connectivity. Ian Buck, NVIDIA's vice president of hyperscale and HPC, frames it as solving the bottleneck that appears after agents propose an experiment: the compute to test it. At the same conference, NVIDIA also expanded its CUDA-X libraries for use by AI agents.

The numbers not published matter just as much

NVIDIA gave no memory capacity or type, no power draw, no double-precision performance, no price, and named no launch partners or customers. For comparison, the current Rubin GPU has 28,672 CUDA cores — Rubin HPC nearly doubles that — but more cores do not automatically mean faster simulation: sparse workloads depend more on memory bandwidth, which is exactly the part not disclosed. A 2028 ship date also makes this a reveal two years early, with key numbers still missing for any buying decision.

Rubin HPC and Vera Rubin's sweep of the MLPerf training benchmarks are two sides of the same platform — Vera Rubin sweeps the MLPerf training leaderboard on debut: single-card performance doubles, scaling efficiency near 97%: one side trains large models, the other runs simulations for research agents. NVIDIA's story is that the same generation of chips should power AI factories and catch the next wave of scientific computing.

Recommended Tools

More