Particle.news

NVIDIA Unveils Rubin HPC With More Than 50,000 CUDA Cores

The company says the GPU combines sparse-data acceleration and seventh‑generation NVLink to keep many cores fed for faster scientific simulations.

Overview

  • NVIDIA announced Rubin HPC on Monday, Nov. 16, 2026, presenting it as a Vera Rubin‑family GPU built to speed simulations and data processing for science, engineering and analytics.
  • NVIDIA says a single Rubin HPC GPU will exceed 50,000 CUDA cores and use sparse‑data acceleration to reduce waiting for scattered values inside large datasets.
  • The company also highlighted seventh‑generation NVLink and a 'balanced design' that scales data movement alongside compute so many cores stay supplied with data during recurring simulation patterns.
  • Independent press analysis of NVIDIA imagery inferred multi‑die Superchip layouts, eight HBM memory sites and server trays that could exceed 200,000 cores, but those system‑level configurations have not been confirmed by NVIDIA.
  • NVIDIA lists Rubin HPC as expected to be available in 2028, and the launch sits alongside Vera CPU and Nemotron for Science initiatives that aim to shorten the simulation loop for researchers while leaving final shipping timing uncertain given past product changes.