Case Study: Worcester Polytechnic Institute accelerates deep learning training with Google for Education

Worcester Polytechnic Institute trains deep learning 7.7x faster and 62.9% cheaper with Google for Education

Researchers at Worcester Polytechnic Institute (WPI) sought to make deep learning training more efficient and affordable. Their challenge was to determine if cheaper, transient cloud servers could be used for distributed training instead of more expensive on-demand options, despite the risk of servers being revoked at any time. They partnered with Google for Education, utilizing Google's pre-emptible VMs on Google Compute Engine for this research.

Google for Education provided the infrastructure for a large-scale empirical study. By using a cluster of Google's pre-emptible VMs, the WPI team achieved results that were up to 7.7 times faster and 62.9% cheaper on average than using a single on-demand server. The study demonstrated that transient servers are a viable and highly cost-effective solution for distributed deep learning training.

Worcester Polytechnic Institute