Learn about a groundbreaking data center efficiency metric in this technical presentation from Microsoft engineers at the Open Compute Project. Explore how the Combined Performance Metric (CPM) offers a more comprehensive approach to measuring data center performance compared to traditional Power Usage Effectiveness (PUE), particularly for AI workloads. Discover how CPM calculates efficiency by considering the total number of GPUs, floating point operations per second per GPU, and grid power consumption. Follow along as the speakers demonstrate CPM's practical application through benchmarking tests comparing different cooling scenarios for GPU servers, providing valuable insights for optimizing modern data center operations in the age of artificial intelligence.
Combined Performance Metric for Data Center Efficiency in the AI Era
Open Compute Project via YouTube
Overview
Syllabus
A Combined Performance Metric for data center efficiency in the AI Era
Taught by
Open Compute Project
Related Courses
-
Infrastructure Utilization Efficiency Metrics for Data Center Sustainability
-
Designing In-House Server Solutions for Modern GPU Cloud Computing
-
Dielectric Cold Plate-based Modular Data Center Solutions for Efficient Computing
-
High Performance Data Center Storage Using DPUs - Optimizing Power and Performance
-
Scaling Down While Scaling Up - Design Choices That Increase Efficiency and Performance
-
Sustainable Strategies for Data Centers in the AI Era
