San Francisco Compute is a company that provides affordable pre-training clusters for startups, grad students, and research labs. They sell compute by the hour rather than by year and offer a flexible scheduler that allows users to burst to large portions of the cluster so they can scale up their model without having to sign an expensive long-term contract. The company currently has 1000+ H100s coming online in the next few months and offers SSH access as well as setup for Slurm/Kubernetes out of the box. Infiniband with 3.2TB/s is also available upon request.

- Reserve compute available capacity at $2.85 GPU/hr
- Contact sales if you have any questions about capacity or capabilities
- Explore job openings in engineering supercomputing/machine learning systems or sales
- Follow San Francisco Compute on Twitter (@sfcompute)