Nvidia Drops Rubin HPC GPU, Science AI Agents and an IBM Quantum Tie-Up at SC26
Nvidia's SC26 keynote in Chicago delivered Nemotron for Science, a Rubin HPC GPU, and a quantum co-design platform built with IBM and Fermilab. Ian Buck also touted a CUDA-Q decoding algorithm that cuts error-correction research time to 45 minutes, down from 12 weeks.
- Original source: blockchain.news.
- The source published it on Nov 18, 2026 at 2:43 AM.
- No second outlet has reported it yet. Treat it as a single-source claim.
Nvidia's vice president of hyperscale and high-performance computing, Ian Buck, took the SC26 stage in Chicago on Tuesday to unpack a stack of research tools: Nemotron for Science, the Rubin HPC GPU, an expanded Vera CPU platform and a quantum collaboration with IBM and Fermilab. Buck called AI agents a turning point for scientific work, saying they have been a game changer in how researchers interface with supercomputers, devices and instruments.
Nemotron for Science bundles the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes for life sciences and chemistry, and can run locally on DGX Spark systems or a lab's own supercomputer. At Argonne National Laboratory, an agent built on it planned and tested hypotheses, ran simulations and surfaced two new binding sites from a single prompt. Nvidia also pointed to an agent that fixed 81% of broken GROMACS setups overnight on a DGX Spark, and another that sped up a weather model routine by more than 5x on Nvidia GPUs.
The Rubin HPC GPU targets scientific, engineering and AI workloads with over 50,000 CUDA cores, three times as many as its Hopper predecessor. It spans precisions from FP64 to FP4, adds sparse-data acceleration and carries seventh-generation NVLink for high-speed connectivity. Nvidia expects Rubin HPC to launch in 2028.
Vera CPUs, which keep AI agents running tool calls between model steps, are getting a wider rollout. Germany's Climate Computing Center, DKRZ, picked Vera CPUs for its upcoming HLRE-5 supercomputer, with deployment slated for early 2028.
On the quantum side, Nvidia is pairing DGX B200 systems with IBM quantum machines, including hardware powered by IBM Quantum Nighthawk, as part of a co-design platform with Fermilab. Buck also introduced a CUDA-Q decoding algorithm called TCBP-JOBS that shrinks error-correction research work that previously ran 12 weeks down to 45 minutes. A simulation on Jülich's JUPITER supercomputer showed a tenfold reduction in the operations needed for one logical step using a technique called magic state cultivation.
Nvidia rounded out the keynote with Photonics co-packaged optics switches, which Buck said are 15x more reliable on bit errors and link flaps, plus expanded CUDA-X libraries. The company framed the batch as fuel for discovery in fields from astrophysics to materials science and biology.
TopicsNVIDIA · Ian Buck · Nemotron for Science · Rubin HPC GPU · IBM · Fermilab · Vera CPU · Argonne National Laboratory
Source: blockchain.news
Written by AI from verified reporting.
More stories
- Tech & AI x.com 1 viewNvidia falls to third-largest company, now $1.3 trillion behind Apple
- Tech & AI channelnewsasia.com Caught by @Fahri 1 viewJudge Clears Google's $10M Buy of Spirit Airlines Worker Data
- Tech & AI x.com 1 viewNvidia Grabs 76% of Server GPU Revenue as Rivals Trail Far Behind
- Tech & AI x.com 1 viewArmstrong: AI needs a market layer to stay decentralized and free

