NVIDIA Publishes 73k Lines Worth Of 3D Header Files For Fermi Through Ampere GPUs

[JPFSanders]

I think expecting Nvidia to completely open-source their driver is unreasonable. If they would release the firmware for older GPU's, have an open-source kernel driver, and give basic header files and documentation more often, that would go a long way to making life better for nouveau. At least they are making some steps in the right direction. It's too little and too late to convince me to buy an Nvidia GPU, but maybe in another 10 years...

Exactly all that we need from NVidia is that they opensource the kernel driver part and allow for firmware redistribution, they can keep their userspace OpenGL/Vulkan/CUDA as proprietary as they want.

As long as there is a proper open source kernel driver with firmware/re-clocking people will eventually work on open source Mesa drivers for Nvidia.

Nobody hates Nvidia for being Nvidia, what people (me included) hate is their ridiculous open source policies.

Happy to see that even if in small steps they seem to be opening up a tiny bit.

[MorrisS.]

Even if you have the addresses and a clue on what values they accept for setting the different voltages and frequencies, how would you determine what numbers are fine and what numbers will overclock and burn the board? With other stuff (like the stuff covered here in the opengl headers) you can trial-and-error things and only risk getting some bad colors or graphic artifacts. With electrical stuff you risk the hardware. And seeing how old some of this hardware is...

Why reclocking is possible on AMD and Intel GPus by open drivers?

[cooperate]

NVIDIA's Open GPU Kernel Driver is for only GeForce RTX 20 "Turing" series and newer

It’s turing and up. So GTX 16 series and newer are supported too.

[JPFSanders]

Why reclocking is possible on AMD and Intel GPus by open drivers?

Two reasons; AMD permits redistribution of the VGA Firmware, Public information available on how to perform that task via open-source kernel driver.

[Guest]

Why reclocking is possible on AMD and Intel GPus by open drivers?

It's apparently legal hush-hush, but I've heard rumors of being able to extract the firmware from NVIDIA's proprietary firmware, and then use it for clocking with nouveau. I'm not sure on specifics.

[adoptedPenguin]

The fans of an OS which has less 2% market share and practically zero modern AAA games continue to show their true colors. If I were NVIDIA I wouldn't bother.
WE USE LINUX! YOU OWE US!
Repeat ad infinitum. This is just ugly.

but Linux has 100% top500 supercomputer marketshare (and HPC makes heavy use of GPUs)

also, for that "2%" minority group: AAA windows games running on Linux great with AMD open source drivers, more problems using closed-source NVIDIA stack (or even closed-source AMD stack for that matter)

[Quackdoc]

I hope this will help get better coverage in nouveau, currently nouveau is hardly usable on android, but with some work a few folk have managed to get it working, would be nice if this can help that out.

[karolherbst]

Even if you have the addresses and a clue on what values they accept for setting the different voltages and frequencies, how would you determine what numbers are fine and what numbers will overclock and burn the board? With other stuff (like the stuff covered here in the opengl headers) you can trial-and-error things and only risk getting some bad colors or graphic artifacts. With electrical stuff you risk the hardware. And seeing how old some of this hardware is...

the vbios literally tells you that, but for that you need to RE the vbios. Anyway, on GPUs being Maxwell2+ we can't do it, because it requires signed firmware. The new firmware released by Nvidia with their open drivers will allow us to do that for Turing+ and we are already working on it, so stay tuned and hopefully soonish we are able to do just that

[karolherbst]

Exactly all that we need from NVidia is that they opensource the kernel driver part and allow for firmware redistribution, they can keep their userspace OpenGL/Vulkan/CUDA as proprietary as they want.

As long as there is a proper open source kernel driver with firmware/re-clocking people will eventually work on open source Mesa drivers for Nvidia.

Nobody hates Nvidia for being Nvidia, what people (me included) hate is their ridiculous open source policies.

Happy to see that even if in small steps they seem to be opening up a tiny bit.

thing is... whatever we already have just keeps working on whatever driver, so we already have that and people could already work on it.

[Quackdoc]

The new firmware released by Nvidia with their open drivers will allow us to do that for Turing+ and we are already working on it, so stay tuned and hopefully soonish we are able to do just that

I've seen a few people are complaining that moving a lot of stuff to firmware will do anything to put a large wrench in the dev efforts for people working on nouveau, to some even claiming it will break some degrees of mesa driver support. any idea of how this may or may not actually effect mesa's nouveau developemnt? for example can we expect some vulkanor gl features not to work? will it be difficult to work with or is it a non issue?