Nvidia Open-Sourced their Linux GPU Kernel Driver!
ฝัง
- เผยแพร่เมื่อ 10 พ.ค. 2022
- It finally happened! Nvidia released the source code for their Linux kernel driver. What exactly does that mean, though?
Support me on Patreon: / geerlingguy
Sponsor me on GitHub: github.com/sponsors/geerlingguy
Merch: redshirtjeff.com
2nd Channel: / geerlingengineering
#Nvidia #OpenSource #Linux
Mentioned in this video:
- Nvidia's announcement: developer.nvidia.com/blog/nvi...
- GitHub repo: github.com/NVIDIA/open-gpu-ke...
- My Pi GPU Video: • It took me TWO YEARS t...
- Phoronix article: www.phoronix.com/scan.php?pag... - วิทยาศาสตร์และเทคโนโลยี
The radeon GPU Driver was initially not made by AMD, it was an reserve engineered effort by independent engineers similar to the nouveau driver. But AMD first opened up the specifications of their GPUs which helped to get the driver to an much better state and then later added some patches themselves to improve it.
They did eventually open up their Vulkan driver, but after radv had already been out for like six months or a year or so. Since then they've been on top of adding new register definitions, etc., for new products in advance of shipping them, which has helped improve the speed at which proper drivers can get written.
as an open source enthusiast and electrical engineer dealing with datasheets and whatnot that's what actually useful for opensource community, as it won't infringe upon AMD's Intellectual Property (the driver code) nor AMD's patent (not reproducing the hardware) nor NDA (the register/ instruction definitions). what i suffered through in my old laptop's GPU is proprietary GPU just magically break and newer driver won't work because the card is not supported by the newer driver. OSS driver will "support" all of them and tell what works and what not works no need to pin version on specific kernel version
I still remember the horror of fglrx which no guide could help and the only option was to reinstall whatever distro I'm running
@@fltfathin Also this is far better than some hacker releasing nvidia's source code, because noveau can't make use of illegally released code, but since they've now released parts under MIT noveau are totally allowed to look at and use it to improve noveau as long they abide by MIT, which is very permissive
revere engineering project so good its officially supported by the original author
Im willing to bet Nvidia did this because a combination of things you mentioned. Additionally the steam deck is making large ripples and when SteamOS is released they can see the writing on the walls. Additionally China is replacing all windows OS with OSS alternatives. Feels like they waited for the absolute last minute inflection point to comply
yeah that makes more sense to me, they don't want to be out of linux gaming market if steam efforts finally succeed, there's big money to loose if linux suddenly catches tracktion
China has been replacing Windows for 4 years now, all that happened is them not being able to follow through with it so they extended the deadline by another 2 years. Although I do agree that they must be seeing Linux picking up some steam.
@@DMSBrian24 good point, also LHR being unlocked and the reported Nvidia hack by the group LAPSUS could also have something to with this open source shift as well. All in all is the perfect storm!
For me, it's because lapSUS
@@bitelaserkhalif clearly Jeff using a graphics card on a Raspberry Pi is the main reason
I never would have expected this!! Holy moly!
The mandatory blobs still requires signed firmware to work, so for the sake of freedom nothing has really changed
Well, Rossmann started Right 2 Repair, you may have started Right 2 Recompile 🤓
Never thought this day would happen. A step in the right direction indeed
a forced step. nvidia is being spanked by AMD on linux and nvidia can not produce x86 CPUs due to licenses. Their only hope to survive is to build CPUs based on ARM or RISC-V and for that, they must support GNU/Linux in a proper way.
that means they had to opensource their drivers to remain competitive.
No. It's a PR stunt and nothing more. Look at how they didn't open-source anything that actually mattered. They just moved the interesting code outside of the kernel so they can open source the boring code and say "look the kernel part is now open source"
This 100% has to do with secure boot and the fact that governments and other organizations that 1. Run Linux 2. Are under strict security requirements 3. Buy billions worth of hardware. To date these orgs have begrudgingly made exceptions for certain use cases but compliance is king and secure boot is a requirement. Red Hat says to NVIDIA, dudes, we sell a compliant OS and your software taints it and puts it out of compliance. What gives? NVIDIA says, so what? Red Hat says look at who is buying your cards and putting our OS in the systems and how much money we both make from that. NVIDIA says, oh I get it now.
Guess they realized a serious problem with rher gpu failing to be found by computers. It must of started recently when people noticed that their gpu failed to be found.
If you do a little digging, you'll find that no, they didn't open source the driver. They've just moved 99% of the source code into a 34MB firmware blob. Apple's GPU firmware blob by comparison is in the order of 400kB. This is just a PR stunt by NVIDIA whilst also giving the finger to kernel developers.
I didn't look into it myself, but if true it seems somewhat unlikely this gets into mainline and of questionable utility even if it does.
@@timseguine2 The only way anything would hit mainline is if the changes were submitted via the nouveau kernel driver. It's very unlikely to hit mainline as-is
Typical for Nvidia...
Typical
Where is the firmware blob? I cloned the repo and couldn't find anything like what you describe, and I can't anywhere it gets downloaded by the build system either
I don't know your work on the pi with graphics cards may have had more sway than you think. I know I saw you specifically mentioned in an article not too long ago. Congratulations BTW. And yes I think Nvidia is finally seeing the writing on the wall. Probably that not being transparent enough can get you into trouble. Lol
I agree 100% with you. But it's actually Red-Shirt Jeff you meant, they look alike.. it's an easy mistake to make ;)
@@MI7DJT he has a twin? how did i not know that?
@@siontheodorus1501 You rarely see them in the same room at the same time.
Now nothing remains but to get Broadcom to open-source their wifi drivers.
And open source the pi bootloader.
so people can enable promiscuous mode on them jajaja XD
Somehow this all feels like they finally talked to a lawyer and did the absolute minimum to comply to GPL
Maybe some big contract supercomputer needed the GPLed source code for their kernel
I agree, nvidia are a bunch of scum bags. Im sure they did it kicking and screaming but knew they were in trouble after all the price gouging and negative things they've done the past 5 years.
The absolute minimum is what they did before: an open source shim that communicates with the closed source "real driver"
They got extorted by the LAPSUS extortion group lol
My guess is SteamOS 3 broke down nvidia.
(My speculation) Valve wants the whole SteamOS as distro to be completely open source so that they make their hardware last longer and other OEMs can just use the iso from Valve's website in their handhelds. Now this compounded with the stuff you mentioned in the video and the fact that Valve's Wayland compositor works well with open source drivers but won't even show up in most cases on Nvidia must have forced Jensen to finally give up and release the driver.
I seriously doubt so. The amount of red tape around this will be incredible. I would not be surprised if this move comes from 2 or 3 years ago.
what would prevent steamOS users from installing a proprietary driver
@@dudelookatree the fact that the GUI and the Proton game compatibility layer don't support NVIDIA binary driver as a 3D acceleration backend. If you have a NVIDIA card you are limited to text console only. See HoloISO project (repackaging of SteamOS 3.0)
@@dudelookatree note that I'm talking of the current SteamOS, not the old and discontinued version that can still be downloaded from the website, afaik that old version had NVIDIA drivers built in
They got ransomwared. Release driver source for Linux or get the source released anyway and DIE CAD files.
It was a far different landscape back in the "Super" VGA days when chip and board makers would send you a board and 2 kgs of printed technical manual to enable driver development.
But I'll have to admit there wasn't really much of a performance advantage. It was more a matter of being able to actually use VGA modes beyond IBM's original spec on specific hardware. No close standardization at the hardware level, so a "Tseng Labs" driver was more of a fixup for usability than for performance.
Man, I couldn't believe it either!! From what I understand, however, NVIDIA can still keep most of the "secret sauce" in a firmware blob that is run in privileged mode on the GPU. So, at this point, there is no downside to them open-sourcing these "outer" portions of the driver. It's truly a win-win situation. They get a driver maintained by more people and can focus their software engineers on the cutting edge stuff. I hope this will also help AMD, and give us more leverage for cross platform efforts for graphics and ML APIs (eg. Vulkan, CUDA)
My dad works at Nvidia and he said it was totally your raspberry pi video that made them do it.
I work at Nvidia and can confirm Geerling is responsible.
Nvidia can't be missing out on that sweet sweet raspberry pi.
Can confirm. I am Nvidia.
i work at NVIDIA and can confirm that it was grace cpu did it ...
Cool
Thanks for the quick upload! I'm really excited to see where this goes, and I *hope* that with enough participation, we can get them to release those user-space modules! This should, at the very least, be a big help to the nouveau project
I really hope this leads to a better and actually fully usable experience for Linux with Nvidia. As ive heard it before, we basically have no idea how the proprietary driver works, so we just have to improvise it to work with Linux, hence all the horrific screen-tearing issues, bugs, compatibility problems, etc. Still, my next GPU is going to be AMD, i'm done fighting with Nvidia's bullcrap. The Nvidia's Wayland incompatibility alone is enough to make that choice easy
Won't hold my breath, and I'd definitely still buy AMD by default right now - they have earned it with their open source support, with really good and efficient hardware, for now anyway. Always time for a company to disappoint.
Still this will hopefully be good news that really develops rather than a flash in the pan, and will get back ported a little further than the current generations only - lots of older Nvidia hardware out there begging to get away from the sluggish bloat that is the latest supported Windoze version and be able to perform probably better than they did new under the version of the Windon't they no doubt shipped with.
wayland is trash so nvidia is based for not supporting it tbh
Nvidia chasing the money, even if it's a small %.
Also if the new encoders are true for RDNA3 then I'm never looking back to Nvidia
it's weird I've had exactly the opposite experience with NVIDIA and AMD under Linux. I've bashed against AMD driver bugs and super weird things that needed kernel compiles to get around but every NVIDIA card has "just worked" aside from before dkmis needing to run the installer again after updates. Any graphical issue or monitor back I've had I've been able to solve with xorg.conf Intel I think still won't allow you to override edid which bites.
still, I'm hopeful this leads to better integration and a smoother experience going forwards
@@zyeborm You are not using one of the now ancient gpu that unfortunately straddle AMDGPU and radeon so are not that well supported by either are you?
If the kernel code supports reclocking, this is really great. Even if they don't release the 3D drivers, I'm sure Nouveau or something like it in Mesa will be able to fill the gap fairly quickly using Gallium. Very exciting!
Wait, this video wasn't released on April first? I'm excited
A 2 minute video made in a day of this announcement, you know it's a good announcement when this happens
Hackers already had it, they threatened to release it to the community, so NVIDIA wanted to look like the good guy and released it themselves.
and they got arrested on the news when fbi and swat came crashing into hacker home...its been over the news lately
I love listening to you talking about tech stuff, learn so much,
Thanks Jeff. I heard so many conflicting opinions about this and you just give us the information so we can make our own opinions. Love this channel.
Thank god with the polar ice cap melting maybe it'll balance things out by the way two very cool uploads recently this and the camera good job
Appreciate the level-headed coverage of this news :-) I have been waiting for this since I first dabbled in Linux 18 years ago. This is like Christmas ... well it's like Christmas and getting a boat. You'll have to wait 6 month to really use it.
Indeed, mighty chilly here all of a sudden
I wonder if at any point they will open source the drivers for the models they no longer support like the
"Linux traditionally has supported obsolete hardware"
Not true in the slightest. Windows has always had far greater out of the box compatibility with older computer hardware precisely because it isn't run by open-source puritans. In my 20+ years of building computers, I haven't encountered one that failed to install Windows while there have been dozens that either completely failed to install Linux or required weeks of additional work to get everything working properly.
It's simply not a desktop OS and anyone trying to use it as such is a masochist or a cultist.
… and sustainability
@@henlofren7321 OK then try getting an ISA soundcard to run on Windows Vista, or an Empia chip based TV or capture card on Windows 7 - I bought one Empia ezcap retail in 2008, and it was no longer compatible with Windows 7 just months later. In 2014 I bought a new fresh mainboard with a VT2021 Vinyl soundchip which sounds great. And worked excellent on up to windows 8.1 but was impossible to get to run adequately under Windows 10 - the new system exposed some bugs in the older driver and the manufacturer completely fumbled providing new ones, like they did but they never ever worked properly with something always broken.
On Linux I have had to fix some drivers myself in the past but at least I could do it!
Then I have a laptop where any driver upgrade on Windows causes broken backlight control! And at least some Linux distros work perfectly fine with it out of the box.
I do have a professional audio interface from Windows Vista era that still runs OK on 10 with minor bugs that weren't present on Vista or 7, so that's a little miraculous, using the Vista driver. But it runs on Linux as well and does fine.
I've had an insane number of "throw dumb shit at it. And it just works" experiences with Linux.
@@henlofren7321 From my experience windows keeps hardware compatibility for about 10 years. Anything older than that will not work. Though things might have changed (I havent used windows since Win7). For linux I was able to use any hardware I needed even stuff that was close to 20 years old. All I had to do is enable a module in kernel (modules for outdated hardware are usually disabled by default).
@@henlofren7321 Windows has greater support for more hardware but that does not count as great for older hardware, a good example is windows 11, planned obsolescence. A lot of people will sell old hardware which won't support windows 11 just because they made tpm 2 mandatory.
Damn you're fast!
This really IS news!
Seems promising for us with old systems using Nvidia cards that want to switch to Linux.
EDIT: By old I mean ~3-4 years. My laptop won’t be supported since it uses a GTX 1050 4GB but my desktop uses a RTX 2080 8GB.
Unfortunately the open source driver only works with newer desktop GeForce cards (like 16xx, 2xxx, and 3xxx).
@@JeffGeerling Thanks for the reply! I run an RTX 2080 card and might eventually switch over to Linux. Seems like a good idea to wait.
@@JeffGeerling PHEW, I was worried there that my somewhat new card (1660 Ti) in my laptop would be among the ones not included. Still, that sucks for those with older cards...
@@JeffGeerling thought as much. (license permitting) hopefully the nouveau team will be able to fork some code for older cards, especially the ones that haven't had official support in years
Hopefully the drivers can be used for older cards too. Scince it's open source, nobody can forbid you to modify the driver.
It was totally your pi video that did it! :D
Just stumbled upon this video "Jeff Geerling" ... geerlingguy ... this is you?! The guy i got so many ansible roles from? :D Nice to meet you and thanks a lot for all your contributions!
Thanks!
WOW! Thanks Jeff. Now I just have to check if hell has indeed frozen over. I use NVIDIA’s GPUs under Ubuntu to encode videos via ffmpeg.
What kind of frames per second and at what resolution can you get with your setup?🤔
I saw this video on my sub feed and just screamed WHAT.
Now Nvidia can make volunteers do work on the drivers for free and increase profits from those data centres. Smart business move!
Hey, if I can get an A100 running on my Raspberry Pi... I'm all for it
@@JeffGeerling "if" being the operative phrase here. The FW blob is huge, so what have they actually released... On closer inspection, not much...
Wow, that seems a surprise!
Definitely steps into the right direction.
Hopefully this means we'll be able to change clock speed some day on newer GPUs.
Ohh i love your T-Shirts!
Well when the source code was hacked I guess they thought screw it.
Holy shit finally, really exited to see where it goes from here
What a scoop! Damn that’s awesome!
Now, this was worthwhile breaking news on Linux. :)
Wow, that was a quick video turnaround.
Looking forward to seeing some cards running on lunix as well as mac os on m1.
Seeing how this is going to allow people to experiment more down the line, we can see that a lot of doors can be opened again in the future E-GPU's can come back in support should companies like apple allow it.
Did not see that coming. I am in awe !!!! WOWWWWW
About damn time!
Just read the title and got excited
wow... this was a short vid, i was some what disappointed I didn't get that feeling of rolling over in bed reading a good novel..... but carry on Jeffrey!
Nothing to say really, this is an awesome first step, just commenting to boost engagement.
this is really good news. i hope support is eventually expanded to older cards, too. the 9 and 10 series are still quite performant and it'd be a great way to extend their lifespan and keep good hardware out of ewaste.
This is an exciting development! Thanks for the reporting on it
So... they did fu. all but throw whatever source they grabbed, and pushed it to a github repository. Probably because some partner required them to gpl the source, or even better - somebody found out they used gpl code in the core driver, which wouldn't surprise me at all.
I knew as soon as I saw the news on Google you were gonna make a video about it lol
Now it's looking great.
have you leave some welcome to Linux messages for NVIDIA in the `Welcome to Linux` issue on the repo?
wow. never thought I'd see that
Thank. God. Holy crap, I never thought I’d see the day!
I am excited to see what this does for Linux hardware people like System76. This is a step closer for people like them to be able to offer a fully open source machine, top to bottom, without giving up any power to closed source machines.
I mean they can already do that (aside from firmware) with AMD, but realistically you can't fully avoid blobs somewhere if you're using anything x86.
@@rpavlik1 Of course, though I feel like System76 does try to avoid as much of that as it can in it's laptops at least, like they ship with the ME disabled on the Intel processors, and ship with custom firmware for the motherboard, and so on.
I'm sure a manufacturer would need to make all of it's own hardware from scratch if they wanted to be 100% open source everything, but this has to be a step in the right direction with the hardware we currently have at least.
@@knightshousegames oh for sure. It's always nice to have alternatives
... I started considering switching to linux on my main computer like, 3 days ago and have been doing research into linux non stop since then, and then 3 days later Nvidia open sources it's drivers... I'm technically an Agnostic Atheist, but, yeah fuck it this is a sign.
lol now you need to research some other proprietary stuff and maybe it'll be like magic, all open source.
My reply to the thumbnail after see video description.
"Indeed”
Did you update your video to add blur to Twitter What's Happening?
Distinctly remember seeing some celebrity garbage over there a couple hours ago
It was those hackers that held them ransom recently, one o the requests was opensourcing their driver
that's a very good news and will help push gamer on Linux even more
1:46 "Cards close to the chest" - pun intended?
hopefully this allows for better legacy support. i have some old hardware but choose to run bleeding edge distro.
Phoronix got the driver built and working. Commercial games in his test suite virtually don't have any significant difference in average fps but benchmarks are where the driver suffers. Still, pretty good for initial release
When I saw the title of this video I thought of Tesla licensing their battery technology to anyone "for free". What a wonderful deal till you read the fine print. That in mind my expectations were very low but happily it appears Nvidia is doing something positive.
It’s not like Tesla has anything to offer. Their cells are basically off-the-shelf from Panasonic, LG and CATL. Sure, they have shiny new formats - that don’t have too many parts that could actually be considered intellectual property, and they have not been able to build them themselves.
@@nanolog522 You're absolutely correct about their current batteries cells being "off the shelf". I believe it was their battery charging technology they were "sharing". Their new battery cells appear to be more of the same in a larger size.
@@jmr totally unrelated but the larger size Tesla cells are the same chemistry yes but the internal construction is different in ways that should equate to more power, less heat and easier assembly. it's not a game changer but it's a decent improvement.
It's a great milestone for the tech world!
Lets see how this will go. I hope it will not end up as the NTFS driver.
Will this make vGPU or GRID support on consumer cards more non-hacky and maybe even allow SR-IOV support? Or is that still locked somewhere else?
No, as all of that is controlled by the blob userspace part
I Think Only Intel Alchemist Cards Will Support SR-IOV In The Market
Jeff, I'm struggling... I need your help!!! I've a EGPU (RTX 3050) setup on my MacBook Pro (intel) and I would like to use CUDA on docker for macOS not only in bootcamp windows or even using and other type of emulated environment like qemu or virtual box... Do you have any tip or light or even voodoo black magic???
i appreciate that but the suffering form the past wont be forgotten
Indeed. This is the start of a long journey. Hopefully one paved with happy developments! But for now I'm still more of an AMD fan.
0:07 You can now recompile the kernel once again
Not just once!
What a time to be alive.
That's humongous news I totally was not expecting. Good on nVidia!
good on nvidia for "open sourcing" the kernel shim and hardly anything important, like the userspace componenets that actually make up the majority of the driver. yayyy.
I'm giving you 100% credit for this
Will take it a long while to get it into the kernel source tree.
This is amazing. It actually gives me a twinge of hope deep in my gullet.
Hell yeah!!!
That's some good news!
I wonder if this will lead to longer support for older GPUs.
"Older" starting from the current generation, yes. This generation will be supported for a long time thanks to the kernel code being updateable. But the user space blob will still suffer compatibility issues.
It's about time, but we still got a few more modules to go
Guess I'll keep my 2060 after whatever upgrade I pick next. It's good to have a backup system.
Can you put a link for the shirt in the description?
It's a small victory, but we take what we can get
It's about time
Actually, it's about Nvidia open sourcing their drivers...
/dadjoke
Oh my god!!!
So it seems like these drivers have now hit Beta status. How well do they work now for all things Linux desktop and gaming?
Miracles do happen!
Any chance of this making the IME opensource?
It would be great if the GT710 would work with this driver, because it is one of the few cards without microcode. So it would be truly open source.
Wait so is Pascal in alpha or they just not touching it at all 🤔 Not sure what was meant by GeForce.
Still holding out hope for something to be done about the vkd3d/dxvk thing Where running DX12 over Vulkan is limited by a problem apparently Nvidia managed to workaround it in Windows but not on Linux. Might've been Bindless Uniform Buffers or some other Vulkan thing that wasn't supported in hardware.
They’re not releasing open source drivers for anything before Turing - likely because the government and enterprise customers pushing nVidia to make this decision have replaced all their pre-Turing hardware already.
GeForce is all their consumer GPU’s. nVidia’s entire Linux focus is datacenter. Consumer desktop is like 0.1% Linux whereas most servers - including everything running AI/ML workloads - are running Linux. They’ll support consumer GeForce products for Linux, but only because 98% of the work is already done with their work on enterprise grade products that generate all the income.
@@benjaminlynch9958 Damn, a part of me thought a lot more people would still be on Pascal because of the GPFlation and all :/
Dumb question:
How does one compile a GPU driver?
I thought the whole point of a GPU is to have a different specialized instruction set from the CPU.
So, if the GPU has a different instuction set, how does a CPU based C compiler emit GPU instructions?
Seems like I am misunderstanding something here.
I knew I felt cold !!!
And not just because I’m visiting my home state of Washington from Arizona… 😝
Oh and yes my brothers kitchen table now looks like a data center..
Brought the learning cluster with me!
I was just thinking today why nvidia can't release open source drivers due to cuda cores and dlss, and this happened. Of course, it's a pity that they are still closed source binaries.
Because they make more money with vendor lock-in. Abolish capitalism.
With the advent of the steam deck a lot more companies either are or will start paying more attention to Linux, win, win for us all in my opinion.
That’s great news
nice.
POG
pretty sure torvalds is currently sipping his beverage of choice chuckling "too little too late"