Intel's AI Flame Graphs software is now open-source. This is a project that started for Intel's Tiber AI Cloud to provide more insight into AI accelerator/GPU usage and hardware profiling of the full software stack. After being an internal/customer-only software project for some months, AI Flame Graphs is now open-source.

AI Flame Graphs is for exposing Intel GPU performance profiles based on hardware sampling and generating visualizations of the execution via flamegraphs. This includes EU stalls, CPU stacks, GPU kernel information, and more, AI Flame Graphs was started around Intel Data Center GPU Max Series needs while has been extended to the Intel Arc B-Series "Battlemage" graphics cards as well as Lunar Lake Xe2 integrated graphics.

Making use of AI Flame Graphs currently requires kernel driver modifications, frame pointers to be enabled for the built code as well as all system libraries, and eBPF support. The iaprof software isn't limited to profiling AI workloads but can also be used for profiling traditional graphics workloads too.

Those wishing to learn more about AI Flame Graphs now as an open-source project can do so via [1]this blog post by Intel engineer Brendan Gregg. AI Flame Graphs is now open-source under an Apache 2.0 license on [2]intel/iaprof via GitHub .



[1] https://www.brendangregg.com/blog/2025-05-01/doom-gpu-flame-graphs.html

[2] https://github.com/intel/iaprof