Case Study: Multinational facial-recognition leader achieves 73% GPU utilization and 2X faster training with NVIDIA Run:ai

How One Company Went from 28% GPU Utilization to 73% With Run:AI

Multinational Company, a world leader in facial recognition and computer vision, was struggling with low and inconsistent on‑prem GPU utilization (about 28%), static allocations, scheduling bottlenecks, and an imminent >$1M hardware purchase to scale capacity. To address these challenges they adopted NVIDIA Run:ai’s Atlas platform to provide pooled, on‑demand GPU access, hardware abstraction, and better visibility across their 24 DGX servers and additional GPU workstations.

NVIDIA Run:ai implemented dynamic, organization‑wide resource sharing, automated GPU allocations, and cluster visibility, which raised average GPU utilization to over 70% (from ~28%), doubled experiments per GPU, and cut training times by roughly 75%, enabling 2X faster model training. These gains removed the need for the planned million‑dollar hardware buy, simplified workflows for 30 researchers across two continents, and made it much easier to scale deep learning across teams.