Optimizing AI/ML workloads for sustainability
summary
Cloud guidance focuses on measurable savings: AWS claims AWS is 3.6x more energy efficient and 88% lower carbon versus median enterprise datacenters, recommends using Graviton3 gives 3x performance for CPU inference, and points to concrete tools — SageMaker Training Compiler, Debugger that auto‑stops buggy jobs, Model Monitor, Inference Recommender, serverless inference and Inf1 instances. Practical tactics include multi‑model endpoints, managed spot training, Parquet storage and Bayesian tuning uses 10x fewer jobs to cut compute, storage and retraining frequency.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:55Cloud sustainability
- 2:30AWS sustainability journey highlights
- 3:53It's more sustainable in the cloud
- 4:30Shared responsibility model for environmental sustainability
- 5:19Well-Architected machine learning lifecycle
- 5:54Business goal identification
- 6:26ML problem framing
- 9:14Data processing
- 13:54Model building
- 18:31Model tuning and evaluation
- 20:09Model training Reduce the volume of logs you keep by setting limited retention time for your storage
- 24:39SageMaker Debugger
- 26:41What happens after the model deployment ?
- 27:47SageMaker Model Monitor
- 29:38Summary
related talks
20mSustainability architecture design for ML workloads
30mGreen AI in Cloud Native Ecosystems: Strategies for Sustainability and Efficiency
20mBuilding GenAI That Doesn't Cost the Earth
45mGreen AI: Making Machine Learning Environmentally Sustainable
19mResource Management Innovation for AI Workloads in Cloud Computing
34mBuilding scalable end-to-end deep learning pipelines in the cloud
27mGreening the AI Cloud: Validating Power Models for Kubernetes Containers
30mCost Transparency and Model ROI