Nvidia unveils its Rubin HPC GPU with "more than 50,000" CUDA cores, up from 28,672 in the current Rubin GPU, for agentic workflows and scientific simulations
First reported by Tomshardware ·
High-performance computing tasks using Nvidia hardware will see significantly faster simulation and processing speeds starting in 2028.
Nvidia has unveiled its new Rubin HPC GPU, designed to accelerate agentic workflows and scientific simulations. Announced at Supercomputing 2026, the Rubin HPC boasts over 50,000 CUDA cores, a significant increase from the current Rubin GPU's 28,672 cores. This substantial boost in processing power suggests a potential multi-die configuration, possibly linking four dies together using NVLink-C2C technology, similar to Nvidia's Rubin Ultra GPU. The chip will feature Nvidia's NVLink 7 interconnect for high-speed data transfer between GPUs within a rack. While specific details regarding HBM capacity, bandwidth, and FP64 compute rebalancing were not disclosed, Nvidia emphasizes its suitability for scientists and engineers working with large datasets, aiming to speed up simulations and accelerate discovery in their respective fields. The Rubin HPC GPU is slated for release in 2028.
The substantial increase in CUDA cores and the implied multi-die architecture for the Rubin HPC GPU signal Nvidia's aggressive push to dominate the high-performance computing (HPC) and artificial intelligence (AI) markets. By focusing on "agentic workflows" and scientific simulations, Nvidia is directly targeting the most computationally intensive tasks, indicating a strategic move to capture a larger share of the burgeoning AI infrastructure market. The inclusion of NVLink 7 further reinforces Nvidia's commitment to creating a robust, high-bandwidth ecosystem for accelerated computing, essential for the complex calculations required in advanced research and development.
This release positions Nvidia to further entrench its market leadership, potentially pressuring competitors to match its architectural innovations and core counts for specialized HPC workloads. The emphasis on agentic workflows suggests that Nvidia foresees a future where AI agents are integral to scientific discovery and complex problem-solving, making its hardware a critical enabler for these advanced applications. The market should watch for how this increased compute power translates into tangible performance gains for specific scientific disciplines and whether other players can offer comparable solutions by 2028.
AI-written summary. May contain errors.