The ACSS GPU servers are shared by many researchers. This collection brings together the practices that help everyone get the most out of them without getting in each other’s way.
Practice sheets
- Sharing the GPU servers Simple rules so that everyone can use the shared GPUs and disks.
- AI Tools Local inference, VRAM sizing, fine-tuning and other AI tools on the servers.
- Distributed & Multi-GPU How to scale your work across multiple GPUs — and when you actually need to.
- Project Guidelines Common rules for code produced on the ACSS compute servers.
User documentation (login required) ↗ Request access to the cluster
Authors
Collection coordinated, checked on the machines and partly written by Bruno Chaves Ferreira (ACSS), building on the original guides by Mouhamed Sy, Olivier Caron and William Aboucaya, with contributions from Antoine Mazières.
Last updated: November 5, 2026