Case Study: a large medical imaging company achieves faster, lower-cost ML training with Persistent Systems and Google Cloud Platform

Persistent compares GCP against AWS for running ML training on GPUs for a medical imaging company

Persistent Systems helped a large medical imaging company that makes portable ultrasound devices evaluate whether Google Cloud Platform could improve the training of hundreds of TensorFlow-based deep learning models compared with AWS. The company wanted faster training, lower run costs, and easier experiment management for its ML workloads.

Persistent Systems set up and ran benchmark training experiments across GCP and AWS using multiple GPU types and machine configurations, including preemptible VMs and Kubeflow Pipelines on GCP. The result was about 30% cost savings on similar hardware moved to GCP, up to 75% savings with preemptible VMs, and an average 18% reduction in training time, while also making deployments easier and demonstrating GCP’s autoscaling capabilities for deep learning training.