Case Study: King’s College London achieves 31× faster experiments and 2.1× higher GPU utilization with NVIDIA Run:ai

London Medical Imaging & AI Centre Speeds Up Research with Run:AI

King’s College London–led London Medical Imaging & Artificial Intelligence Centre for Value Based Healthcare, based at St. Thomas’ Hospital, uses NHS imaging and healthcare data to train deep‑learning models for screening, diagnosis and personalized therapy. The centre faced low and inefficient GPU utilization (often below 30%), overloaded queues where small jobs blocked larger experiments, poor scheduling and limited visibility that slowed research. To fix this, they partnered with NVIDIA Run:ai and deployed the Run:ai platform (Atlas).

NVIDIA Run:ai pooled the centre’s heterogeneous DGX‑1 and DGX‑2 resources, using a two‑layer virtualization to allocate fractional and full GPUs dynamically, provide advanced monitoring, and enforce fair, priority‑aware scheduling so researchers could run many concurrent jobs smoothly. As a result, NVIDIA Run:ai increased average GPU utilization 2.1X, made experiments 31X faster, enabled 1.85X more experiments, and removed bottlenecks with elastic workloads—speeding development and deployment of critical diagnostic tools.

King’s College London

M Jorge Cardoso

Associate Professor