Computer Architecture Podcast Ep 22: Measuring Datacenter Efficiency and Visioning the Future of Computer Architecture with Dr. Babak Falsafi, EPFL
Dec 10, 2025
Dr. Babak Falsafi, a professor at EPFL and president of the Swiss Data Center Efficiency Association, delves into the intricate world of data center efficiency. He highlights challenges in measuring energy flows and contrasts established metrics like PUE with unknown IT energy use. Babak warns against overprovisioning by customers and discusses the benefits of emerging technologies like Rack2.0 for disaggregated computing. Sharing insights on GPU utilization for AI, he emphasizes the importance of co-design in crafting sustainable, efficient architectures for the future.
AI Snips
Chapters
Transcript
Episode notes
Measure Energy Flows End-to-End
- Data center energy must be measured end-to-end to find where electricity is actually lost.
- Babak argues that focusing on measured flows reveals true efficiency opportunities beyond coarse metrics like PUE.
Standardize IT Efficiency KPIs
- Create and adopt standardized IT metrics focused on utilization, PSU losses, and fan power.
- Report these KPIs so operators can compare and reduce electricity wasted in racks and servers.
AI Shifts Energy Toward Data Movement
- AI workloads shift energy from compute to data movement and interconnects.
- Babak highlights memory, storage, and network as key domains for measuring electricity use in AI racks.