Case Study: Synthesia achieves faster AI model training and maximum GPU utilization with WEKA

Synthesia - Customer Case Study

Synthesia, the enterprise AI video communications platform, needed to accelerate AI model training while reducing the manual data management and storage bottlenecks that were slowing research on AWS. Using GPU-powered Amazon EC2 P5 instances, the team faced low GPU utilization, inconsistent data handling, and performance limits with legacy storage as it trained increasingly realistic text-to-video models.

To solve this, Synthesia implemented the WEKA Data Platform on AWS. WEKA’s zero-copy, zero-tuning architecture, automatic tiering between Amazon S3 and local NVMe, and support for Snap to Object helped eliminate manual data movement, fully saturate GPUs, and simplify multi-region data migration. With WEKA, Synthesia improved researcher productivity, increased infrastructure efficiency, and was able to migrate its dataset across AWS Regions with minimal downtime.

Synthesia

Alex Balan

Technical Lead