How AI and Deep Learning could effort the Climate Change
summary
Proposes a practical system that measures per-job power and converts it to CO2 using grid and renewables data, integrates with the workload manager, and enforces CO2 budget enforcement by delaying nonurgent jobs, switching jobs to renewable times or sites, and using checkpoints to avoid restarting long trainings. Describes lightweight monitoring agents on GPUs/accelerators and InfiniBand links, a dashboard for data scientists, and a center-level policy daemon; early trials show lower energy draw and fewer emissions when policies are active.
AI-written summary · about
chapters (6)
- 0:00Introduction to AI and climate
- 3:22Energy impact of AI models
- 10:13Optimizing AI for efficiency
- 17:30AI tools for climate monitoring
- 24:17Future of sustainable AI research
- 28:19Q&A and closing remarks
related talks
22mLet's Terminate Climate Change with AI
45mGreen AI: Making Machine Learning Environmentally Sustainable
57mThe Next Decade of Software Is About Climate - What Is the Role of ML?
25mGreen AI with JobRunr: Sustainable Training and Inference on the JVM
28mAI vs Climate Change: Insights from the Cutting Edge
22mKEYNOTE: How to Leverage Data to Combat Climate Change 🌳
01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory
07mKeynote: Finding Your Power to Accelerate to a Sustainable Future