NVIDIA announced on Monday (November 16), at SC26 in Chicago, a three-part package for scientific computing: the Rubin HPC GPU, a rack-scale platform with 192 Vera CPUs, and Nemotron for Science, a bundle of an open model, a chemistry dataset and post-training recipes for scientific discovery agents.
Rubin HPC: over 50,000 CUDA cores for simulation
Rubin HPC is a new GPU in the Vera Rubin platform, designed to accelerate simulation and data processing for science, engineering and analytics. According to NVIDIA, it has a record number of CUDA cores — more than 50,000 on a single GPU, the company's own claim, with no independent measurement yet — and includes a sparse data acceleration capability, with no figures given. The confirmed interconnect is seventh-generation NVLink; its bandwidth was not disclosed.
NVIDIA positions the GPU for AI agents that run simulations to test candidate materials, molecules and designs, with its CUDA-X libraries providing optimized implementations of the algorithms. Availability is expected in 2028, with no price given. The company also disclosed no FP64, memory, power or measured comparison figures against its current GPUs — this is a roadmap announcement, not hardware ready to buy.
Vera rack: 192 in-house CPUs take on x86
The Vera CPU rack-scale platform combines NVIDIA's first fully custom data center CPU with Quantum-X800 InfiniBand networking. Each Vera CPU has 88 NVIDIA-designed Olympus cores, 176 threads and SOCAMM2 LPDDR5X memory delivering up to 1.2 TB/s of bandwidth. The rack brings together 192 Vera CPUs, about 300 TB of memory and 1 PB of storage in a fully liquid-cooled MGX design, with more than 200 TB/s of aggregate peak memory bandwidth. Each 1U tray integrates eight Vera CPUs, NVMe storage and management, and the platform reuses NVL72 rack components.
In NVIDIA's own tests, Vera delivered 1.9x the performance and 2.4x the performance per watt of x86 systems on OpenFOAM's motorBike Large airflow simulation. The only x86 system named in the company's performance material is a two-socket server with an AMD EPYC 9755. On the open-source Weather Research and Forecasting model, NVIDIA projects gains of 1.6x in performance and 2x in energy efficiency versus x86.
DKRZ, the German Climate Computing Center, has selected Vera CPUs for its HLRE-5 supercomputer, alongside GB200 Grace Blackwell systems and Quantum-2 InfiniBand networking. Built by Bull and expected to be available in early 2028, the system will support global kilometer-scale climate simulations with lower energy consumption. “Scientific simulations need CPUs that can keep large datasets moving and sustain performance as applications scale,” said Ian Buck, NVIDIA's vice president of hyperscale and high-performance computing.
The HPC SDK 26.11, with compilers and libraries optimized for Vera, is available now, as is a public beta of the next-generation Fortran compiler. Vera CPU racks are expected from Dell, HPE, Lenovo and Supermicro beginning in early 2027.
Nemotron for Science: open model and chemistry dataset
Nemotron for Science brings together scientific applications of the open Nemotron 3.5 Super VL model, a new open chemistry dataset and post-training recipes to help researchers build specialized agents for scientific discovery. LitmusBench v0.2 is an open dataset with nearly 50,000 questions for training and evaluating language models on chemistry reasoning. The model and dataset are available now on Hugging Face, and Super VL was optimized to run locally on a 128GB NVIDIA DGX Spark system.
Organizations cited by NVIDIA include Lila Sciences, Terray Therapeutics and Therna, plus the U.S. Department of Energy's Argonne National Laboratory and the Jülich Supercomputing Centre. According to the company, Lila Sciences reached 61% accuracy in quantum-dot synthesis using Nemotron, versus 54% for GPT-6 Astra, with inference costs 10 to 20 times lower — NVIDIA's own figures, not independently verified. An agent built at KTH Royal Institute fixed 81% of broken GROMACS setups in a single overnight run.
Context
The three announcements extend the Vera Rubin platform, which Geekiki covered at the NVL72 system launch — NVIDIA Vera Rubin NVL72: up to 30x more throughput per dollar with only 3% more HBM — and arrive at the same SC26 where Dell presented its AI Factory advances — Dell expands AI Factory with NVIDIA at SC26. The timelines differ: Nemotron for Science is available now, Vera racks arrive in early 2027 and the Rubin HPC GPU only in 2028.
Image: official NVIDIA photo for the Vera CPU rack-scale platform announcement at SC26. Credit: NVIDIA (via GlobeNewswire).
Sources
- GlobeNewswire (via Quiver) — NVIDIA Unveils Vera CPU Rack-Scale Platform for High-Performance Scientific Computing; DKRZ Selects It for HLRE-5 Supercomputer (Nov 16, 2026)
- GlobeNewswire (via Quiver) — NVIDIA Introduces Nemotron for Science to Accelerate Scientific Discovery with Open AI Models, Chemistry Data and Post-Training Tools (Nov 16, 2026)
- ByteOrbital — NVIDIA announces Rubin HPC GPU for science workloads, due in 2028 (Nov 16, 2026)
- AIStockWire — What is Nvidia's Vera CPU? Nvidia (NVDA) takes on x86 chips (Nov 16, 2026)
This article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and sources on the internet. Please consult the original sources to confirm the information.
By Muse