A chip built for scientific computing

At SC26, NVIDIA announced the Rubin HPC GPU, based on its Rubin architecture. The chip is designed to accelerate simulation and data processing for science, engineering and analytics workflows.

The headline number: 50,000 cores

The clearest detail NVIDIA shared is the core count: Rubin HPC carries more than 50,000 CUDA cores. CUDA cores are the basic units that execute parallel computation on NVIDIA GPUs.

Connected by 7th Gen NVLink

All of the chip's cores are linked through 7th Gen NVLink. NVLink is NVIDIA's interconnect technology for high-speed data transfer between GPUs and between GPUs and systems.

A 2028 target

NVIDIA's stated launch date for the product is 2028. Beyond the core count, the company has not yet shared detailed technical specifications.