SMART WORLDJOURNALSubmit Manuscript
← All technology news

Computing

NVIDIA Unveils AI Agent Tools and Next-Gen Hardware to Revolutionize Scientific Computing at SC26

At SC26 in Chicago, NVIDIA VP Ian Buck introduced new AI agent toolkits, the Nemotron model, a high-performance Rubin GPU, and Vera CPUs, advancing simulation, AI, and quantum computing integration for researchers worldwide.

NEWS

NVIDIA’s Ian Buck marked the company’s presence at SC26 by detailing innovations that merge artificial intelligence agents with high-performance scientific computing. These agents can autonomously plan, execute, and verify experiments on researchers’ own hardware, signaling a new era in computational science.

Central to this advancement is Nemotron for Science, a toolkit combining the Nemotron 3.5 Super VL open model with specialized chemistry datasets. Researchers can deploy it locally on NVIDIA DGX Spark systems, workstations, or supercomputers to automate hypothesis testing, as demonstrated by successful agent experiments at Argonne National Laboratory identifying new molecular binding sites.

Buck highlighted real-world AI agent applications, including a KTH Royal Institute of Technology tool repairing GROMACS molecular dynamics software setups, the SETI Institute’s agent autonomously analyzing spectral lines with Doppler shift corrections, and NVIDIA’s Nsight AI Porting Agent converting weather forecasting code to GPUs, achieving over a fivefold speedup.

Recognizing agents’ computational patterns, NVIDIA has expanded its Vera CPU platform to support the complex control loops agents require—running code, tool invocation, and result evaluation. The German Climate Computing Center has already chosen Vera CPUs for its upcoming HLRE-5 supercomputer, reinforcing their suitability for agent workloads.

On the GPU front, NVIDIA revealed the Rubin HPC GPU, integrated into the Vera Rubin platform, targeting scientific, engineering, and AI applications. Supporting a broad range of precisions, advanced sparse acceleration, NVLink connectivity, and over 50,000 CUDA cores, it is poised to offer roughly three times Hopper’s performance upon release in 2028.

NVIDIA also showcased progress in quantum computing, including improved quantum error-correction techniques using magic state cultivation and a new CUDA-Q decoding algorithm reducing computation times from weeks to minutes. Collaborative efforts with IBM and Fermilab aim to co-design quantum systems combining DGX B200 platforms with IBM’s quantum hardware.

Additionally, NVIDIA demonstrated early deployment of its photonics co-packaged optics switches powering Lambda’s infrastructure, delivering enhanced reliability and reduced bit error rates and link flaps, critical for modern high-bandwidth networks supporting AI and HPC workloads.

While these innovations promise to accelerate scientific discovery by combining simulation, AI agents, and quantum computing, challenges remain in fully integrating these technologies and ensuring broad accessibility outside leading research institutions.

Source & further reading

Ian Buck Kicks Off NVIDIA’s Week at SC26 as Science Enters the Age of Agents ↗

Source published 2026-11-17. Claims and limitations should be checked against the original announcement or research.