At the Supercomputing 2026 conference, Nvidia has unveiled a new member of its Rubin GPU family, the Rubin HPC, which boasts “more than 50,000” CUDA cores, significantly increasing its computational power compared to the current Rubin GPU, which has 28,672 CUDA cores.

According to Nvidia, the Rubin HPC is designed to accelerate agentic workflows and scientific simulations, making it a powerful tool for scientists, engineers, and analysts working with large data sets. This latest GPU is positioned to deliver faster simulation performance, thereby enabling quicker discoveries in fields such as climate research, genomic analysis, and financial modeling.

Nvidia claims the Rubin HPC features the highest number of CUDA cores on any of its GPUs, potentially indicating a four die layout similar to the Rubin Ultra GPU, which uses NVLink-C2C to link the dies together. However, the company did not provide additional details on the chip’s physical layout.

To ensure efficient data movement, Nvidia will utilize its NVLink 7 interconnect, which maintains the 3.6 TB/s per-GPU bandwidth of NVLink 6 while expanding the scale up domain. Despite this, Nvidia did not disclose any specific details about the HBM capacity or bandwidth of the Rubin HPC GPU.

In addition to its computational prowess, the Rubin HPC is designed to support agentic workflows, which involve the manipulation of large numbers of agents in complex simulations. The company emphasizes that the balanced design of the GPU, with both enhanced compute resources and improved data movement between nodes, will help process massive data sets more efficiently, minimizing idle compute resources and achieving better utilization.

Nvidia highlights the use of its CUDA-X domain specific libraries, which are optimized for Rubin HPC, to further boost performance in agentic workflows. This will enable researchers and analysts to accelerate the reasoning loop from hypothesis to simulation to refinement, ultimately speeding up the discovery of promising results.

While the Rubin HPC is expected to arrive in 2028, the company has already begun positioning it as a key player in the agentic AI and AI chip market. Its advanced computational capabilities and optimized libraries make it a compelling choice for applications requiring sophisticated simulations and data processing.

For those interested in exploring the potential of the Rubin HPC, Nvidia’s focus on agentic workflows and scientific simulations positions it as a critical tool in the ongoing advancement of AI and computational science.

Source: https://www.tomshardware.com/pc-components/gpus/nvidia-announces-rubin-hpc-gpu-with-record-number-of-cuda-cores-at-supercomputing-2026-new-chip-accelerates-agentic-hpc-workloads-arrives-in-2028

Thinking about building an AI product?

Get in Touch