Nvidia introduced the Rubin HPC GPU, a new member of the Rubin GPU family, at Supercomputing 2026. Designed to accelerate AI agent workflows and large-scale data simulations, the chip features more than 50,000 CUDA cores and uses NVLink 7 for data transfer between GPUs. It is expected to launch in 2028. Nvidia has not disclosed its physical layout, HBM capacity and bandwidth, or whether it has adjusted its FP64 computing resources.
Source:Tom’s Hardware · Read original
Leave a Reply