Using roofline analysis to explain and improve workload performance on a chip: establishing peak compute and memory bandwidth ceilings, placing kernels by arithmetic intensity, and identifying whether a kernel is compute- or bandwidth-bound before optimizing.
No statement on what tools can do has been recorded for this skill yet.
The Hub knows of no registered course, programme or certification for this skill yet.
Note from the research register: Performance architecture or ML/HPC software role; learned from the Williams roofline literature plus hands-on profiling with vendor counters.
Register version 1.