Nvidia unveils Rubin HPC GPU with 50,000 CUDA cores, due 2028
The scientific computing GPU uses 7th-generation NVLink and belongs to Nvidia's Vera Rubin platform; simulation benchmarks are still to come.
Nvidia announced Rubin HPC, a next-generation GPU for scientific simulation and data analysis, at the SC26 supercomputing conference, according to digitaltoday.co.kr. The company plans to supply the hardware in 2028.
Rubin HPC carries more than 50,000 CUDA cores, which Nvidia describes as the most ever in a single GPU. GPU-to-GPU connections use 7th-generation NVLink, and the chip is part of Nvidia's Vera Rubin platform.
The GPU works with CUDA-X libraries, which Nvidia says will let AI agents run simulations and analysis using optimized algorithms.
Ian Buck, Nvidia vice president of the hyperscale and high performance computing segment, said faster simulation could accelerate scientific discovery.
Benchmarks needed to judge the GPU's simulation performance will be released later, so no comparative performance figures are available yet.
Quick answers
When will Nvidia's Rubin HPC be available?
Nvidia plans to supply Rubin HPC in 2028, the company said at SC26.
How many CUDA cores does Rubin HPC have?
More than 50,000 CUDA cores, which Nvidia describes as the most ever in a single GPU.