Synthesia Accelerates AI Text-to-Video Model Training 30x on NVIDIA GPUs via AWS

Overview

AI video company Synthesia moved ML model training for its text-to-video SaaS, Synthesia Studio, from on-premises computers to NVIDIA GPU-powered Amazon EC2 instances (P5/H100, P4/A100, G5/A10G) using PyTorch, Amazon EKS, AWS ParallelCluster and AWS Batch, accelerating model training by 30 times and supporting 456% user growth.

The challenge

After growing rapidly to 350 employees and 50,000 customers, Synthesia found that training ML models on on-premises computers had become inefficient. Its production facilities generate many terabytes of data each week, and with more than 50 ML researchers training large models, the company needed a large, scalable data lake and compute cluster to run multiple generative AI models 24/7.

The solution

Synthesia switched to multi-node compute clusters for distributed ML model training on Amazon EC2 P5 Instances (NVIDIA H100 GPUs) and Amazon EC2 P4 Instances (NVIDIA A100 GPUs), and uses Amazon EC2 G5 Instances (NVIDIA A10G GPUs) for data processing and to optimize video rendering runtime. The company manages compute capacity using Amazon EKS, AWS ParallelCluster and AWS Batch, stores large datasets in Amazon S3, and builds its AI workflows on PyTorch and NVIDIA CUDA for video rendering and inference.

Reported business value

Using these managed services and NVIDIA GPU-powered compute instances, Synthesia reduced ML model training time for smaller voice models from days to hours, and supported a user base growth of 456 percent.