Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore advanced AI performance analysis techniques in this 31-minute conference talk from the Linux Plumbers Conference. Learn how eBPF technology is being leveraged to create comprehensive AI flame graphs that address the urgent need for better AI datacenter cost reduction tools. Discover Intel's open source AI flame graph implementation that combines eBPF for kernel driver instrumentation with Intel's eustall hardware profiler for GPU instruction pointer sampling. Understand how this innovative approach creates flame graphs spanning both CPU code (including OS kernel operations) and GPU/AI accelerator code execution, going beyond traditional AI flame graphs that focus solely on CPU code paths. Gain insights into the technical challenges involved in cross-platform profiling, the underlying mechanisms of this observability tool, and potential future kernel developments to support enhanced AI performance monitoring. Examine how this represents the first step in a broader initiative to bring comprehensive performance analysis tools to AI workloads, building on the success of CPU flame graphs which have over 100 implementations across the industry.
