• Triangles In
  • Number of triangles submitted to the tiling hardware.
• Triangles Input per Draw
  • Average number of triangles input to the tiling hardware per draw.
• Triangles Out
  • Number of triangles output after back-face, off-screen and sub-pixel culling.
• Triangle Ratio
  • Ratio of the number of triangles output over input, note that while culling removes triangles reducing the ratio, clipping may increase the ratio.
• Vertex Sharing
  • Percentage average of the number of transformed vertices being shared across primitives, a higher value indicates more efficient vertex sharing, resulting in more efficient cache utilisation.
• Tiles per Triangle
  • Average number of tiles a triangle touches, this indicates how many tiles need to be processed per primitive. A large value may increase the amount of work required during rasterization.
• Geometry Load
  • Percentage of time the tile binning hardware spent processing primitives.
• HSR Efficiency
  • Percentage of pixels submitted to the Hidden Surface Removal hardware compared to the pixels that were actually shaded. For example a value of 50% means that half of the pixels were rejected before fragment processing, saving valuable processing resources.
• Overdraw
  • Percentage of on-screen pixels that were drawn more than once. For example a value of 0% means that every on-screen pixel was shaded exactly once, while a value of 100% means that every on-screen pixel was shaded exactly twice.
• ISP Pixel Load
  • Percentage of time the rasterization hardware spent processing pixels.
• Z/Stencil Load/Store
  • Percentage of time the rasterization hardware spent loading and storing the Z/Stencil buffer.
• ISP Throughput
  • Percentage of tiles that the rasterization hardware is processing concurrently during a 3D workload, a higher value implies better rasterization HW utilisation. A low value could be the result of the renderpass requiring more memory per pixel due to MRT/PLS setup.
• Pixels Out
  • Number of pixels written by the GPU to main memory.
• PBE Load
  • Percentage of time the pixel back-end spent processing pixels. A high value indicates the application is using high resolution render surface(s).
• Vertex Shader Load
  • Percentage average of the time spent processing vertex shaders.
• Fragment Shader Load
  • Percentage average of the time spent processing fragment shaders.
• Compute Shader Load
  • Percentage average of the time spent processing compute shaders.
• Shaded Vertices
  • Number of vertices shaded by the shader core.
• Shaded Fragments
  • Number of fragments shaded by the shader core.
• Processed Kernels
  • Number of compute kernels processed by the shader core.
• Cycles per Vertex
  • Average number of cycles the shader core spent processing vertices.
• Cycles per Pixel
  • Average number of cycles the shader core spent processing pixels.
• Cycles per Kernel
  • Average number of cycles the shader core spent processing compute kernels.
• Register Overload
  • Percentage of time the shader core was stalled due to temporary register pressure.
• Texturing Unit Active
  • Percentage of time the texturing unit spent processing texture requests.
• Texturing Unit Overload
  • Percentage of time the shader core was stalled waiting for a texture request.
• Memory Interface Load
  • Percentage utilisation of the interface between the GPU and main memory.
• GPU Clock Speed
  • Current GPU clock in Hz. This may change during a profiling session due to DVFS etc.