
The Toronto Raptors migrated their player performance model retraining from on-premises GPUs to AWS, using Amazon SageMaker and EC2 to dramatically accelerate model updates and deliver faster, more accurate data-driven insights for coaching, scouting, and player health management decisions.


<div><p>This article describes how the Toronto Raptors migrated from on-premises GPU infrastructure to AWS to accelerate their player performance model retraining process.</p><ul><li>Raptors use deep learning models for player performance insights across scouting, coaching, and health management</li><li>On-premises GPUs couldn't handle computational demands; retraining took days, limiting updates to weekly or monthly</li><li>Models process 1-2 terabytes of data from 10 years of player tracking across 1,230 games per season</li><li>Model drift requires frequent retraining to keep predictions accurate as player conditions and strategies evolve</li><li>Migration to Amazon SageMaker and GPU-optimized EC2 instances dramatically accelerated retraining speed</li><li>Future expansion includes 4GB per-game full body pose tracking data integration</li></ul><p>AWS enabled the Raptors to retrain models faster and more frequently, delivering timely insights for competitive decision-making in the NBA.</p></div>


Related articles