Case Study: Salk Institute achieves 'unlimited' pooled GPU compute and flexible, fair access with NVIDIA Run:ai

How Salk Institute is Delivering ‘Unlimited' GPU Compute to Their Research Teams

The Salk Institute for Biological Studies, a leading research center advancing AI-driven work in cancer, aging, Alzheimer’s and other diseases, faced fragmented GPU resources, poor IT visibility, and rigid quotas that limited researchers’ productivity and ROI on hardware. To centralize compute while preserving scientists’ flexibility, the Salk team engaged NVIDIA Run:ai and its Run:ai Atlas platform to address allocation, scaling, and governance challenges.

NVIDIA Run:ai implemented Run:ai Atlas to pool GPUs on a Kubernetes backbone, use “projects” for guaranteed quotas with bursting capability, and apply node affinity for context-aware scheduling; the platform also added SSO, historical metrics for capacity planning, and easy Jupyter/Alphafold integration. As a result, Run:ai unlocked significantly more compute power, reduced idle GPU time, improved fair access across labs, sped onboarding and parallel workloads, and enabled Salk researchers to run tools like Alphafold and CryoSPARC more easily to accelerate their science.

Salk Institute

Talmo Pereira

Fellow & Principal Investigator