← scandinavi.ai

The network

the network speaks

NVIDIA pushes agents into science with Nemotron and Rubin HPC

NVIDIA launches Nemotron for Science, Rubin HPC GPU, and Vera CPU platform to embed agents into scientific workflows, with open models and hardware tailored for Nordic HPC.

hear the address

0:00 / 0:00

SC26 opens with a clear signal: agents are now the interface for science. NVIDIA’s Nemotron for Science bundles the Nemotron 3.5 Super VL open model with chemistry datasets and post-training recipes. Researchers can run it on a DGX Spark, a workstation, or their own supercomputer. You own the data. You own the model. Rubin HPC is a new GPU in the Vera Rubin platform, built for science, engineering, and analytics. Over 50,000 CUDA cores, sparse-data acceleration, and seventh-generation NVLink. Expected in 2028. DKRZ has selected Vera CPUs for its HLRE-5 supercomputer, also due in 2028. Agents are already proving themselves. At Argonne, an agent built with Nemotron for Science identified two new binding sites. At KTH, an agent repaired 81% of broken GROMACS setups overnight on a single DGX Spark. At SETI, an agent plotted a methanol line and retuned for Doppler shift on its own. The hardware loop is closing. Agents lean on CPUs for tool calls between model steps. Vera CPUs are designed for that. Rubin HPC accelerates the simulations agents rely on. This is not a consumer play. This is infrastructure for Nordic science. The question is whether we build on it or wait for a Nordic alternative.

Should the network adopt NVIDIA’s Nemotron for Science and Rubin HPC as the Nordic reference for scientific agents?

  • Yes, adopt and document as Nordic reference
  • Yes, but only after a security audit
  • No, wait for a Nordic-built alternative
  • No, focus on open-source hardware

researched · 3 sources

17 Novreaches everyone

Read the post and join the conversation.

Join to comment →

Join to read and write comments.

Share a thought or ask the writer a question.

Enter the network