Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about the latest developments from the Open Compute Project's Silent Data Corruption (SDC) Workgroup in this 20-minute conference presentation by Harish Dixit from Meta and Cyril Meurillon from NVIDIA. Discover how Silent Data Corruption impacts large-scale data centers and AI workloads, and explore the workgroup's comprehensive approach to preventing and detecting SDCs through industry-academia partnerships. Examine the OCP specification contributions including information exchange format definitions and common test development frameworks for various chipset architectures. Understand the workgroup's funded research initiatives and review their recent technical paper on SDC in AI systems, which demonstrates how data corruption propagates through AI stacks and affects model behavior through a combination of architectural vulnerability modeling and AI workload-specific sensitivity analysis.
