2025-01-21 Render & Cycles Meeting

Meetings
1

Attendees

  • Brecht Van Lommel (Blender)
  • Lukas Stockner (Blender)
  • Nikita Sirgienko (Intel)
  • Patrick Mours (NVIDIA)
  • Sahar A. Kashi (AMD)
  • Sergey Sharybin (Blender)
  • Thomas Dinges (Blender)
  • Xavier Hallade
  • Weizhen Huang (Blender)

Notes

  • For Blender 4.4
    • Brecht was looking into all sorts of high and regular severity bug fixes
      • Some of them seem to be coming from drivers
    • Embree update to 4.3
      • See if update to 4.4 is feasible (there are some blocking issues with 4.4)
      • There are two issues:
        • Win11 Embree Cycles crash/memory corruption on Simplify toggle 129596
          • GPU only
        • Cloth animation crash during Cycles CPU rendered viewport playback 131933
          • Happens on CPU, needs to be investigated further
    • Make a TODO for the communication about benchmarking
    • Build: Upgrade DPC++, Level-Zero and Embree 132763
      • Needs review, ideally before Beta so that it goes with the rest of the updates
    • HIP:
      • Windows:
        • Can update to HIP 6.1. It is solid release, but no gfx1200 support
        • For gfx1200 need to have SDK 6.3 which is not available yet
      • Linux:
        • HIP 6.3: Worksations are potentially unstable, but hopefully will be addressed by a drievr update
        • Driver bump is a concerning factor, but hopefullt distros will catch up soon
    • HIP-RT patch:
      • is marked as WIP
      • Code side is done
      • Split into smaller files?
      • Wait for the update of the HIP-RT in upstream
    • OSL memory leak #133107:
      • Driver bug: is fixed in the upcoming driver version. Will be available soon.
    • OIDN update
      • Bump to the build system has been commited
      • Waiting for the libraries to be recompiled
  • Other
    *

Practical Info

This is a weekly video chat meeting for planning and discussion of Blender rendering development. Any contributor (developer, UI/UX designer, writer, …) working on rendering in Blender is welcome to join and add proposed items to the agenda.

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

7 Likes
2

The last time I checked, 6.1 introduced rendering artifacts. Reducing the optimization level fixed this issue without causing a noticable performance regression, but the same fix couldn’t be applied to HIP-RT as it did introduce a large performance regression, and even with it, MNEE had some issues with HIP-RT.

1 Like
3

For the MNEE we’ve decided to possibly allow temporary breakage as it is not very commonly used feature and there is some work we’ve started with Brecht to hopefully reduce the stress on the compiler for MNEE.

It is all unideal, but it would be good to find a way forward, even if it involves some trade-offs.

1 Like
4

In which case, I’d personally recommend 6.2 instead of 6.1 for Windows.

My quick tests suggest 6.2 is more stable/reliable than 6.1. E.g. 6.1 has rendering artifacts with volumes and MNEE, while 6.2 only has issues with MNEE.

5

Oki! Thanks for the input!