Nvidia unveils Rubin HPC GPU with more than 50,000 CUDA cores, due in 2028

Nvidia has unveiled the Rubin HPC graphics processor at the SC26 supercomputing conference in Chicago, a chip designed for scientific simulation and data processing with more than 50,000 CUDA cores on a single GPU, a company record, plus sparse-data acceleration and seventh-generation NVLink; it is expected in 2028.
Nvidia on Monday, November 16, unveiled a new graphics processor, Rubin HPC, at the SC26 supercomputing conference in Chicago. Built for scientific simulation and data processing, a single chip carries more than 50,000 CUDA cores, a record for a single Nvidia GPU, and the chip is expected to become available in 2028.
Built for agentic science: faster simulation, faster discovery
Rubin HPC is a new member of Nvidia's Vera Rubin platform, aimed not at AI training or inference but at accelerating simulation and data-processing workflows in science, engineering and analytics, including astrophysics and materials science. Nvidia says the chip is aimed at agentic research: AI agents generate large numbers of candidate materials, molecules and designs, and simulations then evaluate them and guide the next step. Ian Buck, Nvidia's vice president of hyperscale and high-performance computing, said: "faster simulation means faster discovery."
Sparse-data acceleration and seventh-generation NVLink
Beyond the core count, Rubin HPC adds a sparse-data acceleration capability so the many cores are not left idle when handling the sparse data common in scientific computing, and the chips are linked by seventh-generation NVLink to speed GPU-to-GPU data movement. By comparison, the standard Rubin GPU has about 28,672 CUDA cores, so Rubin HPC has nearly twice as many.
Specs and price not disclosed; science models and quantum work unveiled alongside
Nvidia has not published Rubin HPC's memory capacity or bandwidth, double-precision floating-point throughput, power draw or price, and it has not named launch partners or customers. Other scientific-computing announcements at the show include the Nemotron for Science model package for researchers, new CUDA-X libraries, and Slurm 26.11 scheduling support for quantum processors; Vera CPU racks will be supplied by Dell, Hewlett Packard Enterprise, Lenovo and Supermicro in early 2027. SC26 also marks the 20th anniversary of CUDA.
Sources: NVIDIA official blog, Techstrong, StorageReview.
N環球 
💬 Comments
No comments yet. Be the first!