NVIDIA Transitioning To Official, Open-Source Linux GPU Kernel Driver

[Sonadow]

Open source or not, the current semiconductor shortage means that all graphics cards are still priced through the roof and in low supply.

[blacknova]

So... the only change right now would be that akmod or dkms module will build longer, since you will have to build the whole driver not just the shim. And all issues posed by using akmod and dkms would go away only if repos will provide prebuild tested module for their kernels. But the userspace is still closed.

At least nouveau should prosper with access to nvidia's module source code.

[STiAT]

I would be happy enough to use nouveau or respectively Mesa. Though, that will be a long run short of somebody throwing money / developers behind the effort.

It will gelp nouveau a lot with hardware support, something which was a huge issue due to not having DRM code or hardware documentation.

[ElectricPrism]

I'm going to maintain my skepticism.

Of course I would forgive them AFTER they have put in the necessary work to have a decent open driver and have established a decent short term record with being open source.

Until then I am still 100% AMD as I have been the last 5 years. Nvidia definitely has my attention if they want to impress upon me their worthiness to be considered as an option.

Until then I am 100% strictly open source AMD or Intel (I hear they are entering the dGPU market, their open source record is good.)

[dragonn]

Is there any basis to this? Reads like a conspiracy theory nvidia buyers would make to support the "its not nvidia, its everyone else!" status of play

Don't get me wrong, I am not blaming the kernel devs or either nvidia for that, both sides have some truth in it but no, this is not a conspiracy theory, just google nvidia gpl condom https://www.phoronix.com/scan.php?pa...king-NV-NetGPU

[dragon321]

Finally. I didn't expect them to go fully open source so I'm not really dissappointed that user space is closed source. I believe that it wouldn't make much sense for NVIDIA to give their proprietary technologies as open source. Also I believe that sooner or later this gap will be filled as well. But even without that it's step in the right direction. In their site they mentioned that Nouveau can leverage same firmware used by NVIDIA driver and provide things like clock or thermal management. If it indeed means that Nouveau will be finally able to provide reclocking and proper power management in more recent NVIDIA cards then it's really great news. We might end with something comparable to radeonsi and amdgpu where Nouveau will take care of pre Turing cards and new Nvidia driver will provide support for Turing and later with optional closed source user space providing NVIDIA specific features like CUDA, NVENC, NVDEC etc. I hope it will be true and we will be able to get usable open source driver on NVIDIA cards.

[illwieckz]

Very nice. My guess is that eventually Nouvue (I hate this unspellable name BTW)

May be hard to spell, but it is very easy to pronounce anyway: https://fr.wiktionary.org/wiki/nouveau#Prononciation
Basically noovo, with “noo” somewhat like in “noodles” and ”vo” like in “vocabulary”.

For the reference, while the actual free open source driver for actual devices is called “nouveau” which means “new”, the driver for very old devices is named “vieux” which means “old”. The pronunciation of “vieux” may be harder to describe, it's like “view” but instead of “ew” it's “eu” which is hard to describe to english people without actually pronouncing it. The “x” isn't pronounced at all: https://fr.wiktionary.org/wiki/vieux#Prononciation

So basically “nouveau” was a pun because of the previous driver was named “nv” and it happened that ”nv” is a french abbreviation of “nouveau”, and “vieux” is a pun because of “nouveau”.

I don't know how to pronounce “radeonsi” and if it is meant to be pronounced anyway.

[deusexmachina]

Will be curious to see "how open source" vs AMDGPU. For example, how many binary blobs in firmware. How many potential security issues they could introduce (via system memory access for example). And how about the digital rights management implementations - and can they be disabled.

[arunbupathy]

Was only a matter of time. Just waiting to see how fast people forget their past shenanigans. I'll postpone my celebration until they mainline this and develop their driver in the open together with the community.

[arunbupathy]

I would also add to my last post that I will truly consider nVidia good if they open up the CUDA API/ABIs for everybody else. That way other vendors can provide CUDA implementations, and existing CUDA applications can be used as-is, without jumping through a million hoops.