Nvidia unveils Rubin HPC GPU with over 50,000 CUDA cores at SC26
Nvidia revealed Rubin HPC, a new chip in its Rubin GPU lineup, at the Supercomputing 2026 conference. The company says it packs more than 50,000 CUDA cores, the highest count on any Nvidia GPU to date, and is aimed at accelerating agentic workflows and simulation for scientific and engineering users. Nvidia did not disclose the chip's die configuration, HBM memory specs, or FP64 performance details.
GoKawiil's interpretation of the reporting above, not reported fact.
The lack of disclosed FP64 performance is notable given that AMD and Intel are emphasizing native FP64 capability for their competing HPC chips at the same conference, suggesting Nvidia may be positioning Rubin HPC differently, around AI-driven simulation rather than traditional double-precision computing. The core count increase implies a multi-die packaging approach similar to Rubin Ultra, which could hint at Nvidia's broader strategy of scaling compute through chiplet-style designs across its GPU family. Without bandwidth or capacity figures, it remains unclear how Rubin HPC will actually compare against rivals in real-world HPC workloads.
- Rubin HPC has over 50,000 CUDA cores, the most on any Nvidia GPU
- Nvidia withheld details on die layout, HBM specs, and FP64 support
- Chip targets agentic and simulation workloads rather than traditional HPC precision computing
Source: tomshardware.com — Jeffrey Kampman, 2026-11-16
Published there as: “Nvidia announces Rubin HPC GPU with 'record number of CUDA cores' at Supercomputing 2026”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.