Cloud computing and carbon emissions
summary
Explains measurable claims and tools: a reproduced methodology (now open source as Cloud Carbon Footprint) estimates cloud usage → kWh → CO2 using instance-power lookups and provider APIs; a stable diffusion training example uses 150,000 GPU hours = ~11 tCO2e (Hugging Face) and estimators give ~13 tCO2e; a hardware meter project Kepler maps actual server wattage to containers; moving from on‑prem PUE 1.39 to cloud PUE 1.1 yields ~20% site savings. Recommends concrete tactics: time/space shifting, scale‑to‑zero, resource scheduling and finops/green ops.
AI-written summary · about
chapters (16)
- 0:00Introduction and background
- 1:55Sustainability of IT solutions
- 3:43Energy proportionality in servers
- 7:51Programming language efficiency
- 10:16Carbon awareness and electricity mix
- 15:52Green computing foundations
- 16:12Measuring and estimating emissions
- 20:02Cloud carbon footprint methodology
- 23:15Estimating carbon emissions
- 29:35Reducing emissions with GreenOps
- 32:04Public cloud advantages
- 34:10Resource scheduling and optimization
- 35:33Batch job reorganization case study
- 36:59Optimizing ML training
- 38:23AWS Well-Architected sustainability
- 39:47Q&A on cloud providers and neutrality
related talks
23mDriving cost-effective solutions to reduce carbon footprints
42mUndertaking a GreenOps Approach in the Public Cloud
41mWriting Greener Software Even When You Are Stuck On-Prem
46mMaking Software & Data Architectures Sustainable
50mResponsible computing: how about a survey of measurement tools?
41mToward a carbon-aware cloud
55mCode that Won't Heat the Planet: Engineering Sustainable Software for a Better Future
12mKeynote: Cloud Native Sustainability - How You Can Help Right Now!