2021-11-23 Blender Rendering Meeting

Meetings
1

Attendees

  • Brecht Van Lommel (Blender)
  • Kévin Dietrich (Blender)
  • William Leeson (Blender)
  • Thomas Dinges (Blender)
  • Sayak Biswas (AMD)
  • Patrick Mours (NVIDIA)
  • Michael Jones (Apple)
  • Christophe Hery (Facebook)

Notes

Cycles

  • The 3.0 Release is in a good state. There are no high priority bugs remaining and all HIP issues are fixed. We are waiting on a fix for a performance regression on CUDA with scrambling distance to be merged. The Blender 3.0 build should be released on the 3rd of December. (Brecht)
  • HIP: Working on getting everything up an running on Vega cards. Also fixing the Linux build which is having issues with the shared libraries and trying to upgrade to a newer compiler version.D13324, D13042, D13296 (Sayak)
  • MetalT92212: Would like to move quickly to get a fully working version in the code base as quickly as possible. Currently only 4 test cases are failing. We would also like to get the Metal RT patch in. Looking into using run time compilation to specialise the rendering pipeline which had 5%-10% performance impact. However, compilation can take a long time negating the benefit. We are wondering if Blender is set up to build and test Metal builds. Finally, the Metal implementation currently does not have an lgamma implementation that is usable for licensing reasons and this is currently used by some closures in Cycles so one needs to be found that can be used. D13263, D13241, D13243, D13236. (Micheal)
  • Currently Blender can easily support Metal builds as it currently runs on a M1 laptop so it should easily be able to support testing Metal. (Brecht)
  • Optix/CUDA: Patch for Optix 7.4 and newest SDK has been submitted. Nvidia also looked at pipeline specialisation which also got a 2%-5% performance gain but was not worth it due to the cost of compilation. (Patrick)
  • MNEE patch is ready and will be submitted soon. Added support for an arbitrary number of interfaces. However, The patch needs looking over and currently forces ray tracing and will need some work to integrate it. (Christophe)
  • A patch for fixing a performance regression on CUDA D13042 with scrambling distance is complete and is currently going through testing and will be merged soon. Ray offsetting removal D12954 is now complete and appears to work in all cases. Needs some testing but should go in soon. (William)

Eevee & Viewport

  • Opensubdiv patch D12406 is almost done. Attribute rendering has some bugs reports T93179 (Kevin)

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.

  • Agenda
  • Google Meet
  • Time: Tuesday 5 to 6 PM Amsterdam Time
  • Next Meeting: December 7th, 2021
15 Likes
2

BIG Thank You! For both Blender and AMD developers.

4 Likes
3

Yea, Can’t wait to render refractive caustic with Cycle. This is gonna be fun. :100: :nerd_face:

5 Likes
4

Indeed! I am tired of faking non-physical caustics in cycles!

1 Like
5

The material has good controllability in terms of focal dispersion, but it is really too difficult. I wouldn’t mind if an advanced scattering material asset was given, but if the mnee is not spectralized it still needs some skills to get the trigonometric effect, but of course that one is easy

6

Nope, it’s more about Caustics. Dispersion is another topic and is more related to spectral rendering.

7

So incredibly happy to hear about progress on the Metal front—thanks to all involved!

Will the daily Alpha build be the place to download at some point in the future?

My M1 Max and 2019 Mac Pro are starting to get giddy…

1 Like
8

Yes, you should be able to get your daily builds from here. There is lot of Metal stuff going in right now and the aim is to get it working as quickly as possible, so keep checking :slight_smile:

4 Likes
9

Really excited for the Metal implementation! Can’t wait to test out M1 Max!

10

So, when can we expect some sort of build to test for the Metal support?

11

When it’s ready for testing :slight_smile: It will be announced for sure!

2 Likes
12

cyclesx doesn’t support cpu+gpu rendering together anymore?

13

Yes it does. It’s just not very efficient at the moment. In my case it’s even slower than GPU only.

14

Since it is not efficient to render a picture at the same time, is it possible to render the animation at intervals?

15

It’s possible, if you script it yourself. You could start one Blender session with GPU and one with CPU rendering via command line. Not very elegant though…

16

I think if you need to visualize the working status, then you should improve the current image viewer, at least similar to the c4d kind

17

Saw the metal benchmark and M1 pro seemed to take about 80 seconds to render, that is twice rtx3050ti and quadruple 3060 laptop gpu. Can we expect metal patch to improve in a short period regarding performance?

18

The developer working on the Metal Implementation lists “performance tuning” as something that still needs to be worked on. As for how much performance can be gained via performance tuning, and how quickly we will see the patches for performance tuning, that is dependent on a variety of factors that can’t be accurately predicted.

It’s also possible that the performance results you’ve seen so far were done using different scenes and/or settings when compared to the benchmarks for the other GPUs you’re comparing the M1 Max too.

1 Like
19

Is the MetalRT patch in yet?

20

The kernel side MetalRT patch is currently in Blender Master. To keep a track of Metal development check for changes in this report and the associated patchs/commits: ⚓ T92212 Cycles Metal device

Note: Do not ask for performance numbers or time estimates and such on that report.

3 Likes