
Nvidia announces Rubin HPC GPU with 'record number of CUDA cores' at Supercomputing 2026 — new chip accelerates agentic HPC workloads, arrives in 2028 - Tom's Hardware
Nvidia has announced its new Rubin HPC GPU at Supercomputing 2026, featuring a record number of CUDA cores to accelerate agentic HPC workloads.
The announcement comes despite continuing turmoil elsewhere in the company. Nvidia's new Rubin HPC GPU is positioned as a powerful tool for professionals working with large data sets, promising significant performance improvements for complex computational tasks.
Record-breaking Performance
The chip boasts the highest number of CUDA cores on any of Nvidia's GPUs - "more than 50,000" - which suggests it may use a configuration similar to the four-die Rubin Ultra GPU. These dies are linked together using NVLink-C2C technology to maximize computational power.
Future Impact
The Rubin HPC GPU is specifically designed to accelerate agentic workflows and simulation, targeting scientists, engineers, and analysts who require massive computational resources. The GPU is scheduled to arrive in 2028, positioning Nvidia to maintain its leadership in high-performance computing.