NVIDIA Announces Nemotron for Science, Rubin HPC GPU, and Vera CPU Expansion at SC26
·2026.11.17 23:00
Key point
The new Rubin HPC GPU features over 50,000 CUDA cores, roughly three times the count of Hopper, and is expected in 2028.
Details
- NVIDIA unveiled Nemotron for Science, the Rubin HPC GPU, and an expanded Vera CPU platform for high-performance computing at SC26 in Chicago.
- Nemotron for Science combines the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes, allowing researchers to run agents locally on DGX Spark, workstations, or supercomputers.
- The Rubin HPC GPU supports precisions from FP64 to FP4, adds sparse-data acceleration, and connects via seventh-generation NVIDIA NVLink, with over 50,000 CUDA cores.
- A new quantum computing co-design platform was announced with IBM and Fermi National Accelerator Laboratory, pairing NVIDIA DGX B200 systems with IBM quantum computers including those powered by IBM Quantum Nighthawk.
Agent Performance and Applications
- An agent built with the Nemotron toolkit at Argonne National Laboratory planned and tested hypotheses from a single prompt, identifying two new binding sites.
- A GROMACS agent from KTH Royal Institute of Technology fixed 81% of broken setups in an overnight run on a single DGX Spark.
- NVIDIA's Nsight AI Porting Agent ported a Weather Research and Forecasting (WRF) routine to GPUs, achieving a more than 5x speedup using about 150 million tokens.
Infrastructure and Quantum Computing
- The Vera CPU platform is selected for the HLRE-5 supercomputer at DKRZ, expected in early 2028, joining systems at the Texas Advanced Computing Center.
- A new CUDA-Q decoding algorithm, TCBP-JOBS, cuts error-correction research computation from 12 weeks to 45 minutes and aims to reduce decoding latency from about 300 microseconds to 10.
- NVIDIA Photonics co-packaged optics switches deployed at Lambda show 15x higher reliability on bit errors and link flaps.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.