Nvidia launches Rubin HPC GPU with over 50,000 CUDA cores for science

Covered by 3 sources
Read onSiliconANGLE· 5 min
Nvidia introduced Rubin HPC, a GPU featuring more than 50,000 CUDA cores and enhanced NVLink interconnect, designed to accelerate agent-driven simulations in scientific and engineering research. The GPU aims to speed up workflows that involve generating and evaluating many candidates in materials, molecules, and designs. Nvidia also released new CUDA-X software libraries to support scientific applications, with Rubin HPC expected to ship in 2028.

What happened

NVIDIA introduced a new GPU called Rubin HPC at the SC26 conference on November 16, 2026. This chip is designed for scientific, engineering, and analytics workloads that rely on simulations to evaluate large numbers of candidate materials, molecules, and designs. Rubin HPC features over 50,000 CUDA cores and advanced data acceleration capabilities, aiming to speed up agent-driven exploration and discovery. The GPU is part of the Vera Rubin platform, with Rubin HPC expected to be available in 2028.

Key details

  • Rubin HPC has more than 50,000 CUDA cores on a single chip.
  • The GPU uses seventh-generation NVLink for high-speed communication between GPUs.
  • It is designed to accelerate sparse-matrix and stencil calculations, sorting, and reductions.
  • The chip is expected to be available for purchase in 2028.

Why it matters

Research teams and companies working in science, engineering, and analytics will be able to run larger and faster simulations, allowing them to evaluate more design or material candidates in less time. This can reduce the time and computational resources needed to move from hypothesis to actionable results, potentially lowering costs and speeding up innovation cycles. Organizations planning to adopt Rubin HPC will need to consider its 2028 availability when planning upgrades to their high-performance computing infrastructure.

Also covered by 2 sources


More in Industry & Manufacturing