Stefan implemented a solution for CUDA primary context retaining
New release with CUDA driver and Metal support is expected next week, just in time for 4.1. Using a git tag or hash is fine for us, which means it might be available sooner as we do not have to wait for binaries to be available.
For CUDA and OptiX, only RTX cards are supported. This means it will be ok to use the latest CUDA toolkit 12 for compiling the library, as compatibility with older cards is not possible anyway.
For AMD, Brian mentioned that the AMD library used by OIDN might get updated to support more GPU architectures.
Discussion about how to control when to use GPU for denoising, and the interface of Cycles GPU settings in general. There are bigger things that can be improved here, but concrete plan for this feature is to add a checkbox in the Denoising panel in the properties editor that allows disabling denoising on the GPU. If there is no GPU that supports denoising, it would be grayed out to also communicate when this is unuspported on GPUs.
AMD GPUs
A fix for the issue with multiple viewports and overlays is being worked on, this will be a fix in the Linux driver.
HIP-RT support on Linux is not going to be ready for 4.1. It is waiting for the HIP-RT library to be open sourced.
Sphere lights: Brecht works on an option to get back the old, non physically based falloff for 4.1.
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.
I believe what this note means is that OIDN NVIDIA GPU denoising only works on RTX cards. So for simplicity sake, OIDN will be compiled with CUDA 12.X.
Cycles will likely retain support for some older cards for rendering for a bit longer.
It would be nice to get some clarification on this. According to the OIDN github page, OIDN is supported on Volta, Turing, Ampere, Ada Lovelace, and Hopper GPUs. Some of these aren’t RTX GPUs. So is it better to say that OIDN supports GPUs with tensor cores?
That also brings up another question. Does OIDN support the full Turing lineup? The GTX 16XX GPUs are still Turing, but lack tensor cores, yet the OIDN page doesn’t make a distinction between GTX and RTX Turing, implying these GTX GPUs may have support.
This is probably a question better aimed at the OIDN developers than the Blender developers.