2021-11-9 Blender Rendering Meeting


  • Brecht Van Lommel (Blender)
  • Kévin Dietrich (Blender)
  • William Leeson (Blender)
  • Thomas Dinges (Blender)
  • Brian Savery (AMD)
  • Patrick Mours (NVIDIA)
  • Michael Jones (Apple)



  • Cycles 3.0 Release: Just a few bugs and the HIP changes remain. (Brecht)
  • Work continues on getting HIP support working it looks like it should support RDNA cards. AMD is working to support as many cards as possible. (Brian)
  • Optix de-noiser fixes have been committed D13158. (Brecht)
  • Metal work continues T92212 with a few more patches coming soon including a MetalRT patch. (Michael)
  • Blender 3.1 is now open and we are looking at the patches that just missed 3.0 such as: fish eye camera, light groups, point cloud rendering, ray tracing precision, animation de-noising. (Brecht)
  • MacOS security update has broken HD4000 support this is being looked into T92090.

Eevee & Viewport

  • Work continues on OpenSubdiv with custom attributes.

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: November 23, 2021

Fantastic news, especially for me and my 5700 XT!
Thankyou Brian and Brecht for all the work that has gone into HIP support.


Interesting update, I suppose MetalRT is using GPU cores or does it use ML cores for help?

What about Vega 64 support? It is on par with RDNA 1 and on the slides from a few years ago I see that it had HIP support planned. It works perfectly fine and I am not looking to buying a new RDNA 2 GPU for thousands of dollars just to be able to render with GPU in Blender again, especially with current state of the GPU market. If I am pushed into upgrading to a new GPU it just might be NVIDIA because of the CUDA support, but I did initially go specifically for AMD for a number of reasons.

Although Vega 64 supports HIP, it has not been properly tested and verified in Blender with the new HIP compiler, drivers, and the code seen in Blender and as such it is not officially supported.

AMD and/or the Blender developers are investigating if older GPU architectures like Vega can be supported, or if some code tweaking are needed, but until things have been properly tested on older hardware HIP support appears to be officially limited to RDNA and newer.

Performance may be on par, but the GPU architectures are quite different which can lead to some issues. This is the largest contributor to it requiring further testing or code changes before being officially supported.


I am building Blender 3.1 devel version from source on a daily basis, I have compiled rocm-hip from AUR and so far rendering with HIP fails every time with the same error, because Vega 64 arch is not even enabled in the source. How is anyone testing if old GPUs are just disabled in the code altogether?

So it seems that up until this point the focus is only on the current GPU architectures. If developers have access only to the current gen GPUs, give the community an opportunity to do the testing, isn’t it the benefit of the open source model in the first place?

I agree. However, for now basically it is RDNA forward that is enabled so as to minimise the test footprint and code changes if any that would be required. We are looking into what else would work, so as to support as many cards a possible. If you are interested, you could enable your card and try to figure out what needs fixing (this is all open source after all :slight_smile: ) all help is very welcome.

To expand on @leesonw’s comment, you can enable GPUs dating back many generations by making similar changes to the source code as found in this commit: rB01f39ef89d40

However it should be noted that in the official task keeping track of HIP in Blender, it states that Linux still needs support stabilized and working driver available. So I suspect one of the main issues for why it’s not working properly for you could be the lack of working drivers on Linux at the moment. Or it could be caused by a collection of other issues related to the exact version of the HIP compiler you’re using or related to how your specific GPU architecture handles the different code paths.

Source: ⚓ T91571 Cycles HIP device



Just wondering: has work already begun on the path guiding implementation? Really looking forward to that.


I also agree with oepncl’s recommendation to run the benchmark a second time prevails, the first time took a lot of time to prepare

I don’t think anyone is currently working in path guiding yet but it was in the plan. However, currently there is work on implementing MNEE which provides similar improvements.


Will the HIP support go into the regular consumer / gaming optimised drivers for AMD GPUs or only the “Pro” GPU drivers? Asking on the behalf of folks who like to play games AND work on the same PC. :slight_smile:


Yes. (needs 20 chars)


Although the pro drive game performance is not bad, but for the latest game pro is still relatively slow to get optimized, such as Forza Horizon 5 new and old drive can differ by 20 percent, which is how horrible the number next door 3080-3090 price doubling are less than 20 percent performance improvement (only games), so adrenaline will join hip?

1 Like