HPC@CVC Lessons on Scaling Up: From 1 GPU to 100s | CVC Workshop

HPC@CVC Lessons on Scaling Up: From 1 GPU to 100s | CVC Workshop

🚀 HPC@CVC Workshop

Lessons on Scaling Up: From 1 GPU to 100s

📅 Date: May 22, 2026
🕘 Time: 09:30 – 14:00
📍 Location: CVC Conference Room

👨‍🏫 Speakers

  • Artemis Llabrés
  • Tomàs Ockier
  • Marc Serra Ortega

🧠 Abstract

From one GPU under your desk to hundreds across Europe’s supercomputers, scaling modern AI is a journey through performance, memory, communication, and infrastructure. This session breaks that journey down into practical steps: how to squeeze more throughput from the GPUs you already have, how to identify bottlenecks, how to reason about memory and compute, and how to make informed choices when scaling beyond a single device.

We will move from single-GPU optimization to distributed training with DDP and FSDP, then into multi-node HPC workflows with Slurm and real scaling considerations. The session will close with a practical introduction to EuroHPC: what resources are available, how to apply for access, and how Europe’s supercomputers can be used to train models on hundreds of GPUs.