November 16, 2026 · Tom's Hardware
Nvidia unveils Rubin HPC GPU with over 50,000 CUDA cores
At Supercomputing 2026, Nvidia announced a Rubin HPC GPU for scientific and engineering work, with more than 50,000 CUDA cores aimed at what it calls agentic HPC workloads. The chip is due to arrive in 2028.
Why it matters: A dedicated HPC part shows Nvidia defending scientific computing just as AMD claims over 40% of new TOP500 systems and argues high-precision workloads won't be replaced by AI. It is also an early sign of product direction under interim leadership after Jensen Huang's exit, and the 2028 date gives labs planning supercomputer purchases a long lead time to compare against AMD's roadmap.
Related updates
- US rule blocks Chinese firms from remotely accessing restricted AI chipsNov 16
- China keeps fastest-supercomputer crown as TOP500 sees record-low new entriesNov 16
- Congress moves to tighten AI chip export rules after Nvidia scandalNov 15
- Meituan open-sources 1.6-trillion-parameter agent model trained without NvidiaNov 15