Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about power management and characterization strategies for AI accelerator systems in this 17-minute conference talk from the Open Compute Project. Explore the critical challenges facing high-power AI accelerator GPU systems, racks, and clusters as AI models continue to grow in complexity and power requirements. Discover comprehensive approaches to power virus development, workload evaluation, telemetry and monitoring systems, and test orchestration for power characterization. Examine data analytics methodologies that help understand at-scale impact on system stability and reliability. Gain insights into performance-per-watt analysis techniques and power modeling strategies essential for optimizing AI infrastructure. Understand the importance of industry collaboration in developing better tooling, testing frameworks, and evaluation methods for power and performance assessment in modern AI systems.
