Luis Chamberlain sent out the modules changes today for the Linux 6.6 merge window. Most notable with the modules update is a change that better builds up the defenses against NVIDIA’s proprietary kernel driver from using GPL-only symbols. Or in other words, bits that only true open-source drivers should be utilizing and not proprietary kernel drivers like NVIDIA’s default Linux driver in respecting the original kernel code author’s intent.
Back in 2020 when the original defense was added, NVIDIA recommended avoiding the Linux 5.9 for the time being. They ended up having a supported driver several weeks later. It will be interesting to see this time how long Linux 6.6+ thwarts their kernel driver.
Would they though? They sell their hardware, not their drivers, or am I misunderstanding something about Nvidia?
Is it possible that by revealing their drivers they would also reveal something about their industry designs?
I mean, just building the hardware and letting the community do all the work on drivers for free would be better, if they don’t do it there must be a valid reason I think.
I mean, they make money of selling the hardware from what I understand. Maybe I’m misunderstanding, and that’s the problem. Maybe they make money off the driver’s too.
Their drivers were already leaked, any secrets they were trying to hide are out in the wild, so that point is moot.
Of course I can’t know for sure because the driver is closed-source, but I’d bet that a lot of what makes Nvidia hardware work fast is actually in the driver rather than the hardware itself. Plus, a proprietary driver lets them lock people in to buying their hardware. The company where I work doesn’t use Nvidia software because it buys Nvidia GPUs. It buys Nvidia GPUs because it uses Nvidia software.
I don’t believe that even for a second. Software doesn’t make hardware run faster. It can certainly slow it down. But it doesn’t make it run better.
Of course software can’t exceed the physical limits of the hardware but reaching the physical limits of the hardware is non-trivial, especially for hardware as complex as a modern GPU.
Not really that difficult to use 100% of GPU resources. I’m developing a game right now. It’s not well optimized and uses 100% of GPU resources depending on what I’m doing in the game.
Reaching 100% utilization is simple and entirely under the control of the user. Optimized drivers are for giving that user more computation at 100% utilization.
Oh yes sure, the software make nvidia gpu better, something that probably most of the hundred if not thousand of contributor to the mesa driver and in the list we have amd, intel, collabora, redhat, nouveau, google, valve and many others didn’t see, they were the only one in the entire silicon valley to find this secret sauce to make gpus better with software.
Yes? I’m not saying Mesa as a whole is bad, but Mesa+Nouveau for Nvidia cards is terrible.
(It doesn’t help that Nvidia isn’t exactly cooperative when it comes to supporting open-source developers, but my point that driver development is non-trivial stands.)
Mesa+Nouveau is bad only thanks to nvidia and their signature lock implemented since the 900 series, as even stated by me before:
You’ve convinced me. I still think that secret optimizations are a possibility, but I must concede that there might very well be this sort of lock-in bullshit and nothing else.