Performance model

ROCm Compute Profiler exposes detailed metrics for CDNA™ architecture-based AMD Instinct™ MI-series GPUs and for select RDNA™ architecture-based AMD Ryzen™ APUs with the supported analysis configurations.

Use the following pages for architecture-specific naming, panel layout, and conceptual overview:

• Instinct (CDNA) - AMD CDNA architecture (CDNA-CDNA4): Architecture and data-type tables, top-level CDNA block diagrams, MI-series terminology, and chapters for:

  • System Speed-of-Light

  • Compute unit (CU)

  • L2 cache (TCC)

  • Shader engine (SE)

  • Command processor (CP)

  • References

• Ryzen APU (RDNA) - RDNA3: Architecture, top-level RDNA3 block diagram, gfx115x panels, and chapters for:

  • System Speed-of-Light

  • Shader engine (overview)

    • Workgroup Manager (SPI)

    • Workgroup processor (WGP)

    • GL0 (TCP Vector Cache)

    • GL1

  • GL2 cache

  • Graphics Core Efficiency Arbiter (GCEA)

  • Command processor (CP)

  • Graphics Register Bus Manager (GRBM)

  • References

For practical profiling walkthroughs, see Profiling by example.