All Connected Using 7th Gen NVLink To Accelerate Simulations & Data Processing As Launch Slated For 2028
NVIDIA Rubin HPC is designed as the company’s most powerful solution for science & engineering workloads, offering over 50K cores.
NVIDIA Brings Record-Breaking CUDA Core Count With Rubin HPC, Coming in 2028
Today at SC26, NVIDIA is announcing its Rubin HPC GPU. The GPU is based on the Rubin graphics architecture and features lots of cores that will be dedicated to accelerating simulation and data processing for science, engineering, and analytics workflows.
While NVIDIA hasn’t released a whole lot of details, what they are disclosing is the core count. The Rubin HPC chip will pack over 50,000 cores on a single GPU. NVIDIA states that the GPU will come with sparse data acceleration capabilities to keep the cores fed with data and will utilize 7th Gen NVLink interconnect technology for GPU-to-GPU connectivity.
That balance is what optimizes Rubin HPC for scientific computing. It keeps the GPU’s record number of CUDA cores supplied with data, so they spend their time computing rather than waiting. This accelerates recurring patterns of every simulation workload — from sparse matrix operations to stencil computations to sorting and reductions.
Based on the image shown by NVIDIA, the chip itself features a large GPU die in the middle with a total of eight HBM sites. Now, based on the arrangement of the chip itself, it looks like there are two Rubin GPUs that are made into one using the NVLink interconnect in the middle, and we can also see a large section dedicated to I/O.
Rubin HPC servers will include trays with the latest NVLink switches, & feature two Superchips, each with a Vera CPU and two Rubin HPC chips. Each Superchip will pack over 100K cores, while a tray would pack over 200K cores.

“For agents tackling science and engineering, faster simulation means faster discovery,” said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA. “Rubin HPC accelerates the simulations agents use to test ideas and guide their next steps, helping scientists and engineers explore more possibilities and bring breakthroughs closer.”
For a good while, NVIDIA has been out of the traditional HPC stage and focusing primarily on the AI side. The company has not made any meaningful acceleration in FP64 compute, while AMD has a special variant of the Instinct MI400 series that is going to deliver some record numbers. The HPC move on Rubin seems to be expanding NVIDIA’s portfolio back into traditional data center use cases, which is good to see.
NVIDIA is putting forward a 2028 “expected” availability for the Rubin HPC GPU. This is more than a year’s wait. NVIDIA also announced the Rubin CPX GPU a while back, which has been cancelled, so given the timeline, it would be interesting to see if Rubin HPC makes it to market or not.
https://wccftech.com/nvidia-rubin-hpc-gpu-50000-cuda-cores-2028/amp/?shem=aimgspe,i3ppal,

