NVIDIA officially unveiled the Rubin HPC GPU at SC26, a powerhouse processor built on the new Rubin graphics architecture. This high-performance chip features over 50,000 CUDA cores on a single die and is specifically designed for science and engineering.
The hardware utilizes 7th Gen NVLink technology for connectivity and includes sparse data acceleration. Server trays will house two Superchips, each containing one Vera CPU and two Rubin GPUs, delivering over 200,000 cores per tray across eight HBM sites.
This release marks a shift in NVIDIA's roadmap following the cancellation of the Rubin CPX GPU. Targeted for a 2028 availability date, the Rubin HPC platform aims to redefine analytics workflows by scaling massive core counts through advanced interconnects.