Nvidia reveals Rubin HPC, a GPU with 50,000+ CUDA cores for scientific simulation
The core count offers the first clue to what Nvidia is building. The current Rubin GPU has 224 streaming multiprocessors and 28,672 CUDA cores, so getting past 50,000 would take at least 74% more cores. Read Entire Article
Nvidia is developing a GPU called Rubin HPC, which boasts over 50,000 CUDA cores, aimed at running simulations for AI agents assisting scientists and engineers in testing ideas. Introduced at the Supercomputing 2026 conference, the chip is part of Nvidia's Vera Rubin platform, although it is not expected until 2028. The high core count indicates a possible multi-die design, with each die containing a significant number of CUDA cores.
Nvidia emphasizes a balanced design, optimizing data movement alongside compute to ensure the cores are always busy processing data. This approach addresses the common idling issue in GPUs when waiting for data during heavy computational loads, such as climate modeling, engineering simulation, genomic analysis, and financial risk work.
The GPU employs NVLink 6 for interconnects, providing substantial bandwidth, although specific figures are not disclosed. Memory capacity and double-precision performance remain undisclosed, with Nvidia focusing on agentic workflows and the efficiency of CUDA-X libraries in accelerating simulation routines. These improvements aim to enable faster discovery by allowing AI agents to explore a larger number of candidates quickly, thereby bringing breakthroughs closer.
Written by urgent.news from TechSpot's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.