NVIDIA Transitions Fully Towards Open-Source GPU Kernel Modules | NVIDIA Technical Blog
NVIDIA Transitions Fully Towards Open-Source GPU Kernel Modules | NVIDIA Technical Blog
With the R515 driver, NVIDIA released a set of Linux GPU kernel modules in May 2022 as open source with dual GPL and MIT licensing. The initial release targeted datacenter compute GPUs…
We’re now at a point where transitioning fully to the open-source GPU kernel modules is the right move, and we’re making that change in the upcoming R560 driver release.
You're viewing a single thread.
Is this really as good as it sounds? There's some parts of the article that concern me:
The initial release targeted datacenter compute GPUs
Not every GPU is compatible with the open-source GPU kernel modules.
Is there any chance that this just means only a certain class of GPU's are ever going to support open source, while their mainline desktop GPU's will still be proprietary?
Not trying to spread FUD, but I don't want to get too excited until I know for sure that this means they will support open source drivers on all their future desktop GPU's.
All GPUs released since they came out with the RTX 2000+ line are supported and all new GPUs will most likely have support, especially with this announcement saying they're committed to it. There's a support list on their GitHub and it includes all the weird little things you'd be worried about. Even silly little laptop chips like the new RTX 500 are on it.
I think the only reason they limited GPU support is because the older ones physically don't have the hardware for this approach; they switched to their newer RISC-V "GSP" processors with the RTX line. In the new open module, all of their proprietary "secret sauce" was shoved off to firmware running on that new GSP. Previously, their proprietary kernel module loaded all of that same secret sauce as a gigantic obfuscated blob running on your normal CPU instead. The Windows side of their driver has also been moving towards using the GSP, they even advertised it boosts performance or whatever, and I can believe it.
That said, with this new stuff, the official Nvidia userland portions providing Vulkan/OpenGL/CUDA support and the like are still proprietary. It's still worse than AMD in that regard. But at least it's possible to replace those bits, and Mesa/NVK are working on getting Vulkan up and running (with NVK supposedly getting pretty damn good, and Mesa's OpenGL-on-Vulkan is pretty good too so that's free).
Ah ok, so it's just the kernel part that they're open sourcing, but a proprietary driver will still need to be installed just as before. I knew there had to be a catch.
I guess it's nice that this would help with kernel issues, like graphics breaking when you install a new kernel. But still not quite what I was hoping for.