Joint Keynote Address: Scaling AI Sustainably: An Uncharted Territory
Talkat 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI '24) · 2024-07-10..07-12·
summary
Claims that AI growth shifts emissions from operations to manufacturing, supported by specific measurements: global data‑center compute rises 550× (2010–2018) while energy only +6%, training a 2021 translation model emits 45.2 metric tons CO2 and recommendation training costs ~9× more, Llama 3 uses 3.75× resources versus prior release. Built artifacts include the mlperf DRM benchmark, the mmax performance model (≈65% throughput gains selecting parallelism), the open‑source Carbon Explorer and ACT carbon modeling tool; these run and expose tradeoffs between operational and embodied carbon.
AI-written summary · about
chapters (13)
- 0:00Scaling AI sustainably
- 6:05AI energy demand growth
- 12:30AI carbon footprint lifecycle
- 22:28Investment directions for sustainable AI
- 24:13Optimizing deep learning models
- 29:28LLM training challenges
- 37:20Jevons Paradox in AI
- 39:04Resilient AI training systems
- 41:55Flexible data center workloads
- 44:01Carbon-aware data center design
- 49:25AI for grid decarbonization
- 51:57Reducing AI hardware emissions
- 57:02Sustainable AI future
related talks
01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory
45mGreen AI: Making Machine Learning Environmentally Sustainable
44mThe Invisible Monster: The Carbon Footprint of AI
20mBuilding GenAI That Doesn't Cost the Earth
57mThe Next Decade of Software Is About Climate - What Is the Role of ML?
25mBuilding a Sustainable Future
14mGreener Intelligence: Reducing AI’s Carbon Footprint
43mGenerative AI and Global Warming: co…