NVIDIA Unveils Rubin HPC GPU for Scientific Agents
NVIDIA announced the Rubin HPC GPU at SC26, featuring over 50,000 CUDA cores to accelerate simulations for scientific and engineering agents.
NVIDIA unveiled the Rubin HPC GPU at the SC26 conference, a chip designed to accelerate simulations for agents in science, engineering, and analytics. The processor features more than 50,000 CUDA cores on a single chip, along with sparse data acceleration and seventh-generation NVLink interconnect technology. Ian Buck stated that faster simulation enables faster discovery for these agents. Dan Ernst noted that Rubin HPC focuses on a specific sub-domain of applications, similar to previous specialized hardware. The chip is expected to be available in 2028, though NVIDIA has not disclosed peak memory capacity or bandwidth.
The broader Vera Rubin platform achieved top results in MLPerf Training v6.1 benchmarks. In its first MLPerf Endpoints submission, the platform delivered more than 10x generational gains in throughput per watt for agentic AI. The Vera Rubin NVL72 system provided up to 2x more performance per GPU than the NVIDIA GB300 NVL72 system. On the DeepSeek-V3 benchmark, Vera Rubin NVL72 scaled from 256 to 1,024 GPUs with about 97% efficiency. NVIDIA also ran the DeepSeek-V3 benchmark on 10,240 GPUs using GB300 systems. For the Kimi K3 model, Vera Rubin NVL72 systems deliver more than 10x the throughput per watt of NVIDIA GB300 NVL72.
Sources
- NVIDIA BlogNVIDIA Announces Rubin HPC — A New GPU to Accelerate Agents for Science, Engineering and Analytics ↗
- NVIDIA BlogNVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance ↗
- The RegisterGPU giants throw HPC users some tasty crumbs ↗
Every sentence of this story was checked against facts quoted from these articles. How we make the news
Spotted an error? Tell us
