ComputeLabs Research

NVIDIA unveiled the Rubin HPC GPU, featuring three times Blackwell’s vector cores and more than 50,000 CUDA cores.

· ComputeLabs Research · from the November 16, 2026 edition

NVIDIA introduced Rubin HPC at SC26 in Chicago as a new GPU within the Vera Rubin platform. Its stated applications are simulation and data processing for science, engineering, and analytics; HPC stands for high-performance computing.

HPCwire reports that the GPU has three times Blackwell’s vector cores and more than 50,000 CUDA cores. These figures describe resources within a GPU, not numbers of GPUs, servers, or racks, and they are not themselves an application-performance benchmark.

HPCwire characterizes the announcement as a renewed commitment to FP64, or double-precision 64-bit floating-point computation, after NVIDIA’s previous two data-center GPU generations did not increase that capability. NVIDIA’s announcement also emphasizes accelerated data movement, but the supplied excerpts do not provide exact FP64 throughput, power consumption, pricing, or delivery timing.

  • NVIDIA
  • Rubin HPC GPU
  • Blackwell

All 20 stories from November 16, 2026