Scaling Massive LLMs

15.04.2026

Moving from a single-GPU setup to a multi-node distributed system is a major technical hurdle. Recently, Fernando Vallecillos Ruiz, a researcher at Simula working on the ImproveIT Project, needed to run DeepSeek V3—a massive 671-billion parameter architecture.

Abstract concept

Running a model this large required a complex multi-node distributed setup using vLLM on our HPC system, Olivia.

Enter our Extended User Support (EUS) and NRIS AI/ML expert Binod Baniya.
Working closely with Fernando’s team, he helped to:

  • Configure a multi-node architecture on Olivia
  • Optimise GPU utilisation across the network
  • Ensure smooth and reliable inference

The result?

A successfully scaled multi-node setup – unlocking the power of massive LLMs for their research

Facing complex bottlenecks?

Let our NRIS experts provide the hands-on support you need through EUS.💡