2024-07-01 EEVEE/Viewport Module Meeting

Meetings
1

Practical Info

This is a weekly video chat meeting for planning and discussion of Blender EEVEE/viewport module development. Any contributor (developer, UI/UX designer, writer, …) working on EEVEE/viewport in Blender is welcome to join.

For users and other interested parties, we ask to read the meeting notes instead so that the meeting can remain focused.

  • Google Meet
  • Next Meeting: July 8, 2023, 11:30 AM to 12:00 PM Amsterdam Time (Your local time: 2024-07-08T09:30:00Z → 2024-07-08T10:00:00Z)

Attendees

  • Clement
  • James
  • Jeroen
  • Miguel

EEVEE

  • Mostly bug-fixing and platform support.
  • Worked around two NVIDIA driver bugs. Hopefully would fix the shadows/light issues we have in the test cases. This will be tested.

Metal

  • Paralize compilation has been worked on
  • Prioritize to Intel/AMD issues.
  • Intel issue has been narrowed down to ray generation. When looking via a debugger to the generated rays it seems to be not what is expected.
  • AMD issue has been narrowed down to a single function in the depth of field. In the debugger this function provides correct results when looking at the work group results, But when collecting the work thread results it only gives one good result per 32 threads, leading to black pixels in the other threads.

Vulkan

The Vulkan backend slowly progresses. It has already proven its benefit when debugging platform specific issues in EEVEE. Some issues detected via the validation layers have been discussed with Vulkan SDK to improve the error messages. For example to use existing debug information inside the shaders to improve the messages to reduce the time when hunting down what is actually going on.

Currently the Vulkan Backend can render Workbench/Grease pencil/EEVEE scenes upto medium complexity. For higher complexity there are still issues that make Blender crash. Cycles is also supported; Cuda/Optix is still untested and most likely doesn’t work due to its different memory model. A short term solution is to download the result and pass it to Vulkan in the same way HIP/OneAPI does.

The Vulkan backend still crashes when playing back animation and doing anything in a background thread (material preview, final image rendering, baking). Performance depends on the scene. The backend still has large room for improving its performance.

15 Likes