PIX is far from finished: we have a long list of things we’d like to improve and add.  This list is both incomplete and uncommitted, so no guarantee that we’ll actually ever do all these!  But we wanted to share it anyway, to give you some idea of where we are likely to go next, and to get your feedback about priorities and what we are missing.  Items toward the top of this list are likely to appear in PIX sooner than the ones further down.

  • Add more Dr. PIX experiments, API usage warnings, and System Monitor counters.
  • Graph counter values along the GPU timeline.
  • Roll up GPU counter values to ExecuteIndirect and ExecuteBundle calls.
  • Shader debugging for DXIL.
  • Tools to help with debugging GPU hangs (aka. TDR).
  • Tools to better understand video memory usage and paging.
  • Tools to artificially simulate running on a GPU with less video memory, TDR, and display changes such as monitor add/remove.
  • Timing captures should show GPU signals/fences and CPU thread names.
  • Better support for the ‘bindless’ GPU resource access model (displaying what resources are used even when a shader selects them dynamically from an unbounded descriptor heap).
  • API summary statistics (how many draw calls, resource barriers, etc.)
  • Support multiple GPUs (PIX currently always just runs analysis on the primary GPU).
  • Support launcher processes (where the entrypoint process launches another, and it’s the second one that PIX ought to capture from).
  • Support for optionally collect Performance Monitoring Counters in Function Summary and Callgraph captures.
  • Assembly level instruction tracing to help you identify candidates for micro-optimizations within a function.
  • The ability to take captures programmatically.
Skip to main content