
NVIDIA Rubin HPC GPU Packs Over 50,000 CUDA Cores, All Connected Using 7th Gen NVLink To Accelerate Simulations & Data Processing As Launch Slated For 2028 - Wccftech
NVIDIA has announced its Rubin HPC GPU, featuring over 50,000 CUDA cores and 7th Gen NVLink technology, with availability expected in 2028.
At SC26, NVIDIA revealed its Rubin HPC GPU, based on the Rubin graphics architecture and featuring over 50,000 cores dedicated to accelerating simulation and data processing for science, engineering, and analytics workflows. The GPU will utilize 7th Gen NVLink interconnect technology for GPU-to-GPU connectivity and include sparse data acceleration capabilities to keep the cores fed with data.
Technical Specifications
Based on images shown by NVIDIA, the Rubin HPC chip features a large GPU die in the middle with eight HBM sites. The design appears to incorporate two Rubin GPUs connected via NVLink interconnect, with a significant section dedicated to I/O operations. Rubin HPC servers will include trays with the latest NVLink switches and feature two Superchips, each containing a Vera CPU and two Rubin HPC chips.
This configuration optimizes the GPU's record number of CUDA cores by ensuring they spend time computing rather than waiting for data, accelerating recurring patterns in simulation workloads from sparse matrix operations to stencil computations.
Market Context
NVIDIA has recently focused primarily on AI rather than traditional high-performance computing (HPC), with limited advancements in FP64 compute. Meanwhile, AMD has developed a special variant of its Instinct MI400 series. The Rubin HPC announcement represents NVIDIA's expansion back into traditional data center use cases.
Timeline Concerns
NVIDIA has set an expected availability date of 2028 for the Rubin HPC GPU, which is more than a year away. The company previously announced the Rubin CPX GPU, which was later cancelled. Given this timeline, questions remain about whether the Rubin HPC will ultimately reach the market.