True nvidia is pretty good on linux... AND Then came wayland.... Fuckkkkkk me on desktop it seems fine but on a laptop with mux switch or optimus it break very badly fucking wayland
and not even 1% of that is because of NVIDIA's efforts, it's all the community gathering together tired of the bullshit of not being able to choose your own OS if you have a NVIDIA GPU.
Trust me, Hell freezes over every year, even the train station which has "Gods Expedition" situated there as well. Of course I am talking about Hell, Norway.
Except for when things suddenly break for no apparent reason like Optix support on my machine a couple days ago. Cuda still works fine, (╯°□°）╯︵ ┻━┻
@@NJ-wb1cz /rant While true, optimized pipelines like Optix *are* Cuda, they're just reified and take advantage of hardware specific functionality for job specific compute. This translates to an average actual speed-up of 30%+ , literally the difference between an Arc 770 and an RTX 4060 for example. And with an Arc it would "just work", which might save the loss in time futzing with this crap. And Nvidia *does* care about these optimized workflows, they just don't care about it working on the desktop. If Intel creeps up on Nvidia for any reason it will be this lack of care, but they aren't stupid. It might be the reason they have started investing more in open source, since AMD doesn't seem interested in making HIP usable.
A murderer who has unconsensual relations with the wife does not "keepalive" the family. Until you slap two sentences of thought together to form an idea I look upon a philosophical Zombie prowling the comment section? Empty handed one-liner comments like yours that posit pandering, placation and cope while attempting to slide us a Null hand. Yeah you want these brains with your slobbering kneejerk reaction, Zombie? You are only the comment that survived not the better man, remember that thieving Zombie. Has my commentary provided you a large enough argumentative attack vector yet? Proprietary does not exist without the good of Open Source, Redhat is just tending the garden thereof. You are a Zombie, because not only do you never question anything you think those who do as lesser than?
@@cameronbosch1213tomato, tomata. Regardless of the name, it's still 1 company that is really doing most of the work in regard to making the system actually WORK.
The most promising scenario is a common kernel driver that can support the Nvidia proprietary userspace as well as open source NVK. CUDA is Nvidia's crown jewels, I don't see that going open any time soon, but being able to use it alongside NVK would be fantastic as it would mean we could have a fully open graphics stack while remaining compatible with all the GPU compute stuff.
CUDA's already half-way separate from the proprietary NVIDIA driver as is. So upstreaming the kernel side gives us a chance to connect it with the existing FOSS userspace stack and at least allow the FOSS and proprietary halves to coexist under a single core kernel driver with optional extra kernel modules, FOSS or otherwise. Basically, this will make it easier to develop a hybrid driver solution that could theoretically run all these feature sets (VFIO, host compute and host DRM/KMS rendering) at the same time.
I'd just be happy to not randomly have kernel updates black screen on boot. Just updated to the 6.9.4-100.fc39 kernel and modesetting isn't working again, forcing me to enter my LUKS password in the dark. Other than this, I've not really had any problems with the proprietary driver in years. The big problems have been the delays in new cards working without it in order for Live USB installers to work at all.
This is exactly what I'd like to see. Let me use Mesa for general desktop usage so I don't have to deal with dumb NVIDIA driver quirks on Wayland etc, but also let me use the proprietary OGL/Vulkan drivers and CUDA for specific apps / games on-demand when I need them, without rebooting. That would be ideal. I don't even care if their official userspace drivers and the firmware remain closed source forever in that case. Just like with AMD and their AMDGPU-PRO userspace drivers that I used to use for Davinci Resolve and nothing else.
Also consider that Nvidia GPUs are also used in supercomputers, and Red Hat is particularly interested in commercial servers, not workstations. But supercomputers almost exclusively run Linux.
When KDE Plasma 6.1 comes out in the Arch repos and the same thing happens with the proprietary Nvidia 555 drivers in the Arch repos, I can see Nvidia GPUs being a much better solution than the past decade or so. Of course, I really wish (wishful thinking but still) that GTX 9 & 10 series reclocking support can be reenabled when they fall out of support with the propietary drivers. That would really get ALL of my old devices off of Windows 10.
To be fair, since the 555 drivers still support GTX 900/1000, those cards will still get the most important Wayland fixes and such, and in the past NVIDIA has updated their legacy driver branches to fix compatibility with newer kernels. So there's hope you will be able to continue using those older cards for many years to come even when the latest drivers drop support and without open source drivers. That said, of course better open source driver support for the older cards would be nice. For Maxwell and Pascal that seems unlikely with the reclocking, but even the older generations which technically support reclocking are still quite miserable on Nouveau compared to the official legacy NVIDIA drivers.
Considering how many people are completely fed up with W11 Nvidia really needs to get stable behavior for their drivers for alternative OSes. As it is, most of the PCs I'm managing right now have AMD cards.
I wish NVIDIA would just open source everything. I find it dubious that is some magic in the code that non-nvidia hardware could benefit from. Their product is hardware.
At the very least I wish they would do something about the firmware situation on pre-Turing cards before/when they drop support in the proprietary drivers so the FOSS drivers can do re-clocking so they don't become completely unusable.
4:30 Can confirm. My GTX 1060 is all I really need. Though with the gpu being this old I am experiencing some graphical glitches sometimes but it's not common. The only feature I would be missing out on would be AV1 encoding
Are you running Linux? I've read that there are some issues when not using the proprietary driver. Like missing blobs to exit low power mode. Can you confirm? This is the only reason why I'm still on windows on my main machine.
What are your thoughts on the xdg-toplevel-icon? Apparently it got merged recently. Any idea on when it's going to arrive to different DEs and applications? I have a few apps that would benefit from this.
4:30 As a proud 1070 user, they definitely perform well enough still. I even game a little here and there. I see no reason to swap any time soon. I hate the feeling of being left behind for what seems like no good reason, or at least a very arbitrary reason.
I care more about getting older GPUs working, people only ever used NVIDIA back then and having all of those stuck on an old outdated driver is awful especially because linux is advertised to revive old pcs
Yeah, pascal and maxwell 2.0 (aka most maxwell cards sans gtx 750 (ti) and some mobile cards - i.e 900 and 1000 series, and also volta but no one has those) is going to be completely screwed on linux once nvidia drops support for them in the prorietarary drivers since they can't be re-clocked. The older cards are not great with open source drivers either but at least they can be re-clocked manually and maxwell 1.0 and kepler could get some vulkan support via NVK if we are lucky.
The 1650 I got used was actually cheaper than the AMD options I could justify spending money on. Nice to see drivers are going to get better right as I got it.
5 месяцев назад+3
The problem for me and others on the data science/machine learning fields, we are kinda dependent on CUDA and there's no alternative for AMD GPUs. Specially for my case, I work and do research in deep learning using CUDA accelerated libraries like PyTorch and Tensor flow, and if you factor how expensive high end GPUs, you only have one option. I really wish AMD stepped up on their game and came up with accelerating tensor operations on their own hardware, and better yet, open source.
Do they want to upstream vGPU stuff? This has always been locked down in the proprietary driver behind a very expensive license and only Quadro cards (even though the hardware is the same on consumer cards and it was possible to unlock it by patching the driver). Also, while not stable, you can already do a vGPU + rendering on the host on 20 series cards and older by merging the vGPU driver files with the desktop driver files plus a few binary patches. With 30 series and up they switched to SR-IOV and broke existing setups
Ugh......So in my attempt to move to Linux(Fedora) AND game I have a 3090FE. The closed source Nvidia drivers have been an outright nightmare. Consistently it refuses to get the correct resolution for my Projector resulting in no video out, and I have to SSH into the system and manually set it.....which was mostly easy until now with Wayland the process it 100% different. Which I'm not certain how to do it in Wayland.) Vs the Framework 16 and its discrete AMD GPU which has been mostly painless making me assume the issue was with Nvidia's crap drivers. To the point that I was looking at upgrading to a AMD GPU. But if things are changing I may hold off on dropping almost a grand on a new GPU. (If I want to replace a GPU that is better than the 3090 on AMD's end it is NOT cheap.)
"Just buy AMD." Yup, that's what I did. Though, that was slightly more about the new PCIe power connector being dog water and wanting to avoid that fire hazard.
@@night_fiend6 No, 16xx are low end 20xx. They're both based on the Turing architecture. 10xx are based on the Pascal architecture. But yes, the numbering was a bit dumb.
@@smithwillnot I actually got my grandmother's old laptop when she bought an iPad. Was my first own computer. When the CMOS battery dies it didn't turn on anymore and I thought it was completely dead, so I got rid of it except taking out the HDD first. Years later I found out that many computers play dead if they don't have power on the CMOS battery, so I was angry at myself and bought a same-model laptop. Yeah, I'm crazy like that. I think it cost me a little above 100€.
Also this needs to happen on the amd side is they really have to allow their decoders so things like handbrake and svp players transcoding can actually make use of the full Hardware. AMF an the other frame generation tech. I’m not saying they have to do some crazy development but they just have to allow it so other people can make use of it because they’re not doing it right now.
Well... I bought my nvidia card and struggled the last months with nvidia-wayland flickering just to be able to use CUDA (I don't really use it for gaming), so unless there is CUDA support from FOSS driver side (which I highly doubt, that nvidia will go that route) I'll have to stick to the proprietary drivers... (or unless there's an viable alternative to CUDA for ML stuff)
As the world's most valuable company, I would not be surprised if Nvidia launches their own brand of high end computers, running their own optimized version of Linux.
There's no reason for a custom distro at all, which is why they don't make one. If you already make and package drivers it's nonsensical to separate yourself from the distros you are supporting.
@@orbatos they wouldn’t be making a distro from scratch, but using one to build a base from. NVidia has branding power to go behind it. Linux does not. A major giant backing Linux could actually bring forth more quality software development for the platform. Competition is desperately needed to break up the duopoly of Windows and macOS.
Yes, I'm with you there on going fully Open-Sourced. We'll just have to see though how this may work out. I just like NVIDIAs internal designs Over ATI/AMD. I'm still using older GeForce and NVIDIAs proprietary drivers 470 version works well there and of course getting the (air quote) ""Manually"" either through CLI or package manager, the supported CUDA drivers as well for my V-Card. I use that for Distributed Computing projects.
Their motivation is gpu splitting functionality for vgpu. Hopefully that functionality makes it to RTX cards, but that's unlikely. Probably targeting mid level business space
@@randomtalkingmicrophone No way since 1998. I've tried linux 3 times at least since I got a pc. Only now, in 2024 have I had zero urge to go back to windows. I dual booted and haven't launched windows once. Might actually delete that partition since it's on a separate SSD.
One of the problems i have, that is usually tied to my Nvidia gpu. Imagin how good it would if they where open source on linux. they could have shifted the market period. I have a gpu, that function 50-60% of what i payed for in many instances. Then I cant use some of it featurs. Im like ready for linux, but its allways this bad feeling you get, sitting there as a gamer now days. yes games works, but you dont get the full gpu effect and performance. Its bad for the consumers that dont follow the norm. The worst part for me is it works, but it does not work as intended. That drives me crazy
It could be a big win-win. They could have a massive, free workforce maintaining and improving the base driver, while offering Nvidia approved snapshots of that driver plus additional closed-source driver components for enterprise customers. The current state just boils down to typical upper-management mentality: Every piece of information given away equals financial loss ^^
Hey Brodie, could you make a video about how Nvidia differs and has been different compared to AMD? I have only heard that "AMD is being good for open source". But I don't know in what way and how. Like what are they doing different? How does it compare to nvidia? And how much better are they actually?
I doubt that they will ever open source their current driver. To my knowledge, such an opening of a proprietary GPU driver stack has never actually happened. The proprietary stack is potentially legally encumbered by third party licenses and such that would render it un-openable. What I hope to happen is that NVIDIA will get more involved with NVK development and help write the new open source driver and get it to a point where it reaches performance and compatibility parity with the existing proprietary driver while also having better integration with Mesa, Wayland, and the Linux ecosystem as a whole than the existing proprietary driver. It seems that there is at least some interest from NVIDIA on this front as their own engineers have done some NVK work. This actually feels like a similar route to what ATI/AMD went down 15 years ago.
I'm a 1080Ti user, I dual boot (mainly use windows for games and Linux for everything else), and I've been thinking of getting an AMD card next because of its superior Linux compatibility. Everyone except Linux users are talking me out of it. Nvidia recently seemingly embracing the open source seems like another good reason But I'll still get AMD. Nvidia didn't do anything yet, and I won't base my purchase on what might happen
If they were opensourcing it on their own public git repo, we could totally merge request some optimisation from some wizard that give a 10k time increase in a specific library use or something.
nvidia is seeing that if they play their cards right, amd may be kicked entirely out of the graphics card space or possibly they see windows tanking eventually given the gross mismanagement at microsoft the right things are happening but i doubt they're for the right reasons
I truthfully wish Nvidia would go full open source and the code being mostly C or C++, put into Mesa. I've been trying and trying to compile NVK 32-bit on my 64-bit LFS system to no avail as it's written in Rust which heavily complicates things with multiple failure points. I have to be upfront in my book saying that compiling lib32-NVK is a useless endeavor on LFS. Nvidia going open source code really change this. Otherwise without the help of another distro that does things in a weird way like Arch and Gentoo (yes it's all strange when examining it all), then the only real option is the proprietary NVIDIA driver which sucks. And no not using LFS is not a great solution. Everything else works. Shouldn't quit because of one chokepoint. Hopefully things get better from here. This might seem like a hot take.
once i said to Jensen: "make thing work" and he opened up vscode and started to cook right in front of me on his apple watch you all are welcome and i may continue buying nvidia not because i'm a 3d artist and i'm forced to, but because it will make the lifes of us all nvidia users better in the long term
I run CachyOS and the devs of Cachy decided the 555 drivers were usable enough and made them available; I installed cachy about 1.5 weeks ago and the install installed the 555 drivers. I've been running plasma wayland with my laptop that has an nvidia 4050 mobile GPU absolutely FLAWLESSLY and have since fully made the switch.
I have been using Nvidia graphics cards for about 20 years and not once had an issue with the proprietary drivers in Linux. Nouveau was, is, and always will be broken. They were broken in 2008, when I first tried nv, and are broken today. OSS for the sake of OSS is a waste of resources.
Sometimes, it's nice to dream that Nvidia is trying to make setting up their GPUs on Linux easier in preparation for the Year of the Linux Desktop. Okay, dream over. Get back to work.
nVidia has seen the self destruction of Windows by Microsoft, and doing what a business does to stay in business. This also is something that is happening already behind the scenes. Their recent move from a Graphic Card company to an AI server company means they are probably using Linux WAY more than any other OS, so this is inevitably going to trickle back into consumer drivers.
Dumb question, but what motivation does Nvidia have to not open source their drivers? Would going FOSS make it easier for AMD or Intel to compete with them? That seems unlikely to me since they've basically got a monopoly on the hardware needed for AI stuff. Assuming it costs nothing to go open source and it improves Linux support, I don't see any reason why they wouldn't do that
At this point dedicating more employees to drastically improve the quality would be a rounding error for Nvidia, they are doing the minimum. At least things are improving.
Nvidia 555 driver still suck on my GTX 2080Ti + Fedora system. Crashes and corrupted gpu memory all the time. 535 remains the best on my machine which forces me to use X11
What about the hdmi 2.1 thing? Remember AMD can't implement it because it's a closed spec. If nvidia becomes open-source will they also lose hdmi 2.1? That'd actually be bad because many gamers need hdmi 2.1.
@felixfourcolor ok, but if it's primarily needed by gamers that is a non issue right since next to no one plays games on their pc hooked up to a TV, and all monitors high spec enough where hdmi 2.1 makes a difference will also have displayport (or usb with displayport alt mode)
@@araarathisyomama787 systemd-tmpfiles manages more that just temporary files so if you tell it to purge them all, at least on Debian it will delete your ~
Next month will mark 4 years of using Linux on Nvidia without issues. Grub causes more issues than Nvidia drivers, imo. Someday there will be a Zluda or Rcom that can compete with Cuda, but until then Nvidia will be the overpriced king.
nVidia is getting there... slowly but surely. And I'd buy AMD, except AMD is always behind nVidia. I'm not going to purchase an inferior product just on principle, I need that performance.
I miss when I could be optimistic about things. Before it became evident that everything any company does will eventually end up being used against us.
" just buy AMD" Is kind of a b******* thing to say to the people who require Nvidia technologies but want to use Linux. And "just by AMD" is why Nvidia isn't putting as much effort into this as they should be I guess what I'm saying is there's a reason Nvidia is a trillion dollar company and AMD isn't despite also being in the CPU market.
@@hegemaniac4652 If you would have actually read my comment, you would have understood that there is a lot of people, myself included who can't just buy AMD because they use Nvidia specific technology nearly daily. [Quite frankly, the lack of drivers for Nvidia graphics cards is one of their only reasons. I continue to use Windows at this point]
It's a joke lol, NVIDIA is the most valuable company in the world not because they're a GPU company, but because they're a GPU company in the middle of AI bubble and AMD completely dropped the ball with compute
Here i have a hot take: since they are selling so many chips with the ai thing, and open ai server rooms are just gpu, maaaaybe it's easier to mass deploy modules, instead of compiling their own kernel.
if you're not developing the driver, why the fuck do you care that it's written in rust? even if you find the community around it annoying, (i do too) that's of zero consequence to you as a user
To be fair, Nvidia is getting a lot easier on Linux the past 2-3 years. Probably very much linked to the fact AI is pretty much 100% Linux.
See Ai is good for one something other than getting companys stealing our data, and jobs. At least we get better Nvidia support for linux :)
True nvidia is pretty good on linux... AND Then came wayland.... Fuckkkkkk me on desktop it seems fine but on a laptop with mux switch or optimus it break very badly fucking wayland
and not even 1% of that is because of NVIDIA's efforts, it's all the community gathering together tired of the bullshit of not being able to choose your own OS if you have a NVIDIA GPU.
@@startjim123456 Wayland on Nvidia actually works pretty well for me since around driver 545 or so. Including offloading of rendering etc.
@@VallThyo Even the comecrial driver gets better Wayland support, and litterally beggs for better vGPU support.
Every time Brodie reads a paragraph twice i feel like there's a glitch in the matrix
"Huh, déjà vu."
His tone is exact same .the binary for his voice would be the same.When it happens I think my RUclips broke
Fcg his 🎾
Oh it's been a while since I missed one of those
Who deleted my comment
Someone should check with the weather service in Hell, because it may have frozen over
Trust me, Hell freezes over every year, even the train station which has "Gods Expedition" situated there as well.
Of course I am talking about Hell, Norway.
@@CMDRSweeper That implies Hell thaws every year in Norway. Hmm...
As an Nvidia user, things are definitely getting better.
Except for when things suddenly break for no apparent reason like Optix support on my machine a couple days ago. Cuda still works fine, (╯°□°）╯︵ ┻━┻
@@orbatos cuda is the main thing nvidia actually cares about
@@NJ-wb1cz /rant
While true, optimized pipelines like Optix *are* Cuda, they're just reified and take advantage of hardware specific functionality for job specific compute. This translates to an average actual speed-up of 30%+ , literally the difference between an Arc 770 and an RTX 4060 for example. And with an Arc it would "just work", which might save the loss in time futzing with this crap. And Nvidia *does* care about these optimized workflows, they just don't care about it working on the desktop.
If Intel creeps up on Nvidia for any reason it will be this lack of care, but they aren't stupid. It might be the reason they have started investing more in open source, since AMD doesn't seem interested in making HIP usable.
@@orbatos ah, I had no idea what it is, sry :)
Reminder how much of the ecosystem Redhat keeps alive.
You mean IBM. Red Hat is Red Hat only in name...
A murderer who has unconsensual relations with the wife does not "keepalive" the family. Until you slap two sentences of thought together to form an idea I look upon a philosophical Zombie prowling the comment section?
Empty handed one-liner comments like yours that posit pandering, placation and cope while attempting to slide us a Null hand. Yeah you want these brains with your slobbering kneejerk reaction, Zombie?
You are only the comment that survived not the better man, remember that thieving Zombie. Has my commentary provided you a large enough argumentative attack vector yet?
Proprietary does not exist without the good of Open Source, Redhat is just tending the garden thereof. You are a Zombie, because not only do you never question anything you think those who do as lesser than?
@@cameronbosch1213tomato, tomata. Regardless of the name, it's still 1 company that is really doing most of the work in regard to making the system actually WORK.
The most promising scenario is a common kernel driver that can support the Nvidia proprietary userspace as well as open source NVK.
CUDA is Nvidia's crown jewels, I don't see that going open any time soon, but being able to use it alongside NVK would be fantastic as it would mean we could have a fully open graphics stack while remaining compatible with all the GPU compute stuff.
CUDA's already half-way separate from the proprietary NVIDIA driver as is.
So upstreaming the kernel side gives us a chance to connect it with the existing FOSS userspace stack and at least allow the FOSS and proprietary halves to coexist under a single core kernel driver with optional extra kernel modules, FOSS or otherwise.
Basically, this will make it easier to develop a hybrid driver solution that could theoretically run all these feature sets (VFIO, host compute and host DRM/KMS rendering) at the same time.
CUDA is the 10-year lead Nvidia has on AMD/Apple/Intel/Google/etc. That moat is - according to the stock market - worth trillions.
I'd just be happy to not randomly have kernel updates black screen on boot. Just updated to the 6.9.4-100.fc39 kernel and modesetting isn't working again, forcing me to enter my LUKS password in the dark. Other than this, I've not really had any problems with the proprietary driver in years. The big problems have been the delays in new cards working without it in order for Live USB installers to work at all.
Isnt that how AMD works on linux? The kernel driver is open source but for userspace you can use mesa or proprietary.
This is exactly what I'd like to see. Let me use Mesa for general desktop usage so I don't have to deal with dumb NVIDIA driver quirks on Wayland etc, but also let me use the proprietary OGL/Vulkan drivers and CUDA for specific apps / games on-demand when I need them, without rebooting.
That would be ideal. I don't even care if their official userspace drivers and the firmware remain closed source forever in that case.
Just like with AMD and their AMDGPU-PRO userspace drivers that I used to use for Davinci Resolve and nothing else.
Starting with the upcoming 560 driver Nvidia recommends the open kernel modules on Turing or newer and will make it the default in their installer.
the steam deck litterally changeed everything. for the better. and is still working its magic down stream way to goo nvidia.
Also consider that Nvidia GPUs are also used in supercomputers, and Red Hat is particularly interested in commercial servers, not workstations. But supercomputers almost exclusively run Linux.
The top 500 supercomputers _all_ run some form of the Linux kernel.
When KDE Plasma 6.1 comes out in the Arch repos and the same thing happens with the proprietary Nvidia 555 drivers in the Arch repos, I can see Nvidia GPUs being a much better solution than the past decade or so.
Of course, I really wish (wishful thinking but still) that GTX 9 & 10 series reclocking support can be reenabled when they fall out of support with the propietary drivers. That would really get ALL of my old devices off of Windows 10.
It is already in extra-testing, using it with the 555 drivers are an amazing experience compared to before, my performance issues with KDE disappeared
Do we have an eta, for when they drop?
To be fair, since the 555 drivers still support GTX 900/1000, those cards will still get the most important Wayland fixes and such, and in the past NVIDIA has updated their legacy driver branches to fix compatibility with newer kernels. So there's hope you will be able to continue using those older cards for many years to come even when the latest drivers drop support and without open source drivers.
That said, of course better open source driver support for the older cards would be nice. For Maxwell and Pascal that seems unlikely with the reclocking, but even the older generations which technically support reclocking are still quite miserable on Nouveau compared to the official legacy NVIDIA drivers.
Considering how many people are completely fed up with W11 Nvidia really needs to get stable behavior for their drivers for alternative OSes. As it is, most of the PCs I'm managing right now have AMD cards.
I wish NVIDIA would just open source everything. I find it dubious that is some magic in the code that non-nvidia hardware could benefit from. Their product is hardware.
The HDMI forum probably wouldn't like that
Software in the only big advantage they have over AMD and Intel.
At the very least I wish they would do something about the firmware situation on pre-Turing cards before/when they drop support in the proprietary drivers so the FOSS drivers can do re-clocking so they don't become completely unusable.
@BrodieRobertson They can quit using HDMI though.
@@dnb5661I really hope they don't do it, I need it for displaying stuff
I wonder if there'll be a point where cuda could be run as a blob on top of a stable novo release, in your NVK and zink environment ✨
4:30 Can confirm. My GTX 1060 is all I really need. Though with the gpu being this old I am experiencing some graphical glitches sometimes but it's not common. The only feature I would be missing out on would be AV1 encoding
Are you running Linux? I've read that there are some issues when not using the proprietary driver. Like missing blobs to exit low power mode.
Can you confirm? This is the only reason why I'm still on windows on my main machine.
What are your thoughts on the xdg-toplevel-icon? Apparently it got merged recently. Any idea on when it's going to arrive to different DEs and applications? I have a few apps that would benefit from this.
I'm glad they are doing something, I may them not like them as a company but this can only make linux better with more competition on linux
4:30 As a proud 1070 user, they definitely perform well enough still. I even game a little here and there. I see no reason to swap any time soon. I hate the feeling of being left behind for what seems like no good reason, or at least a very arbitrary reason.
EVGA 1060 6GB since 2017ish
I care more about getting older GPUs working, people only ever used NVIDIA back then and having all of those stuck on an old outdated driver is awful especially because linux is advertised to revive old pcs
Yeah, pascal and maxwell 2.0 (aka most maxwell cards sans gtx 750 (ti) and some mobile cards - i.e 900 and 1000 series, and also volta but no one has those) is going to be completely screwed on linux once nvidia drops support for them in the prorietarary drivers since they can't be re-clocked. The older cards are not great with open source drivers either but at least they can be re-clocked manually and maxwell 1.0 and kepler could get some vulkan support via NVK if we are lucky.
The 1650 I got used was actually cheaper than the AMD options I could justify spending money on. Nice to see drivers are going to get better right as I got it.
The problem for me and others on the data science/machine learning fields, we are kinda dependent on CUDA and there's no alternative for AMD GPUs.
Specially for my case, I work and do research in deep learning using CUDA accelerated libraries like PyTorch and Tensor flow, and if you factor how expensive high end GPUs, you only have one option. I really wish AMD stepped up on their game and came up with accelerating tensor operations on their own hardware, and better yet, open source.
Wait, that Ben Skeggs video was a couple months ago?
I thought it was a couple weeks ago
Do they want to upstream vGPU stuff? This has always been locked down in the proprietary driver behind a very expensive license and only Quadro cards (even though the hardware is the same on consumer cards and it was possible to unlock it by patching the driver). Also, while not stable, you can already do a vGPU + rendering on the host on 20 series cards and older by merging the vGPU driver files with the desktop driver files plus a few binary patches. With 30 series and up they switched to SR-IOV and broke existing setups
Not fully, I think your'll find the locked down proprietary part will be in userspace and firmware still...
Understandable, have a nice day.
Ugh......So in my attempt to move to Linux(Fedora) AND game I have a 3090FE. The closed source Nvidia drivers have been an outright nightmare. Consistently it refuses to get the correct resolution for my Projector resulting in no video out, and I have to SSH into the system and manually set it.....which was mostly easy until now with Wayland the process it 100% different. Which I'm not certain how to do it in Wayland.) Vs the Framework 16 and its discrete AMD GPU which has been mostly painless making me assume the issue was with Nvidia's crap drivers. To the point that I was looking at upgrading to a AMD GPU. But if things are changing I may hold off on dropping almost a grand on a new GPU. (If I want to replace a GPU that is better than the 3090 on AMD's end it is NOT cheap.)
"Just buy AMD." Yup, that's what I did. Though, that was slightly more about the new PCIe power connector being dog water and wanting to avoid that fire hazard.
2000 series, is GSP.
And 16XX as well, because Nvidia.
@genstian That is what he said in the video.
@@atemoc And 16 comes after 20, because that makes sense.
@@atemoc Yes, 1600 is later than 2000, the lack of logic in numbering is sad. If second digit is 6, you add one to the first digit or something.
That's because the 16xx are basically just low spec 10xx.
@@night_fiend6 No, 16xx are low end 20xx. They're both based on the Turing architecture. 10xx are based on the Pascal architecture. But yes, the numbering was a bit dumb.
11:59 Well, I almost never buy hardware myself, I just take my godfather's old gaming rig whenever he buys a new one.
I read "godfather" as "grandfather", and was like: you must be really young or your grandpa must be really cool.
@@smithwillnot same lol
@@smithwillnot I actually got my grandmother's old laptop when she bought an iPad. Was my first own computer. When the CMOS battery dies it didn't turn on anymore and I thought it was completely dead, so I got rid of it except taking out the HDD first. Years later I found out that many computers play dead if they don't have power on the CMOS battery, so I was angry at myself and bought a same-model laptop. Yeah, I'm crazy like that. I think it cost me a little above 100€.
My EVGA 980 Ti is still kicking butt.
Those 900 cards really punched over their weight.
@@nobodyimportant7804 Yeah unless you owned a 970.
Also this needs to happen on the amd side is they really have to allow their decoders so things like handbrake and svp players transcoding can actually make use of the full Hardware.
AMF an the other frame generation tech. I’m not saying they have to do some crazy development but they just have to allow it so other people can make use of it because they’re not doing it right now.
Well... I bought my nvidia card and struggled the last months with nvidia-wayland flickering just to be able to use CUDA (I don't really use it for gaming), so unless there is CUDA support from FOSS driver side (which I highly doubt, that nvidia will go that route) I'll have to stick to the proprietary drivers... (or unless there's an viable alternative to CUDA for ML stuff)
As the world's most valuable company, I would not be surprised if Nvidia launches their own brand of high end computers, running their own optimized version of Linux.
They are already doing this, apart from a custom Linux distro.
There's no money in the consumer space.
There's no reason for a custom distro at all, which is why they don't make one. If you already make and package drivers it's nonsensical to separate yourself from the distros you are supporting.
@@BGraves they won’t be targeting consumers - for that very reason. They should focus on high end workstations for gamers, creatives and enthusiasts.
@@orbatos they wouldn’t be making a distro from scratch, but using one to build a base from. NVidia has branding power to go behind it. Linux does not. A major giant backing Linux could actually bring forth more quality software development for the platform. Competition is desperately needed to break up the duopoly of Windows and macOS.
Yes, I'm with you there on going fully Open-Sourced. We'll just have to see though how this may work out. I just like NVIDIAs internal designs Over ATI/AMD. I'm still using older GeForce and NVIDIAs proprietary drivers 470 version works well there and of course getting the (air quote) ""Manually"" either through CLI or package manager, the supported CUDA drivers as well for my V-Card. I use that for Distributed Computing projects.
Seems like an editing mistake slipped in
It would be nice if Nvidia just open sourced everything needed for their cards... but I'm not going to hold my breath.
i have plans to do gaming on qubes at some point. this would be very useful for that
Their motivation is gpu splitting functionality for vgpu. Hopefully that functionality makes it to RTX cards, but that's unlikely. Probably targeting mid level business space
damn, 2024 feels like the year of the Linux desktop
has been since 1998!
@@randomtalkingmicrophone Same for me ! 🙂
This seems like from an Nvidia perspective mostly driven by AI and cloud.
@@randomtalkingmicrophone No way since 1998. I've tried linux 3 times at least since I got a pc. Only now, in 2024 have I had zero urge to go back to windows. I dual booted and haven't launched windows once. Might actually delete that partition since it's on a separate SSD.
"finally, x year is the year of the linux desktop" (linux users have been schizophrenically hallucinating this for the past 15+ years)
One of the problems i have, that is usually tied to my Nvidia gpu. Imagin how good it would if they where open source on linux. they could have shifted the market period. I have a gpu, that function 50-60% of what i payed for in many instances. Then I cant use some of it featurs. Im like ready for linux, but its allways this bad feeling you get, sitting there as a gamer now days. yes games works, but you dont get the full gpu effect and performance. Its bad for the consumers that dont follow the norm. The worst part for me is it works, but it does not work as intended. That drives me crazy
It could be a big win-win.
They could have a massive, free workforce maintaining and improving the base driver, while offering Nvidia approved snapshots of that driver plus additional closed-source driver components for enterprise customers.
The current state just boils down to typical upper-management mentality: Every piece of information given away equals financial loss ^^
I dunno man, I haven't had issues with my Nvidia driver on mint, granted I have 1000 series gpu.
10 series is the last pre GSP GPU series
Good for you.
Still waiting for that DLSS Frame Generation support NVIDIA.... still waiting...
At least we have Refelx now... the first stepping stone to FG
New-voh. As in not-old-voh. C'mon, this has been pointed out before.
Hey Brodie, could you make a video about how Nvidia differs and has been different compared to AMD? I have only heard that "AMD is being good for open source". But I don't know in what way and how. Like what are they doing different? How does it compare to nvidia? And how much better are they actually?
how about a standard dmm gpu isa?
i don't want crumbs. Give us the damn cake.
I doubt that they will ever open source their current driver. To my knowledge, such an opening of a proprietary GPU driver stack has never actually happened. The proprietary stack is potentially legally encumbered by third party licenses and such that would render it un-openable. What I hope to happen is that NVIDIA will get more involved with NVK development and help write the new open source driver and get it to a point where it reaches performance and compatibility parity with the existing proprietary driver while also having better integration with Mesa, Wayland, and the Linux ecosystem as a whole than the existing proprietary driver. It seems that there is at least some interest from NVIDIA on this front as their own engineers have done some NVK work. This actually feels like a similar route to what ATI/AMD went down 15 years ago.
Not touching nvidia until the bubble bursts
I'm a 1080Ti user, I dual boot (mainly use windows for games and Linux for everything else), and I've been thinking of getting an AMD card next because of its superior Linux compatibility. Everyone except Linux users are talking me out of it. Nvidia recently seemingly embracing the open source seems like another good reason
But I'll still get AMD. Nvidia didn't do anything yet, and I won't base my purchase on what might happen
i was honestly scared when i heard about x11, but nvidia on wayland has been ok for me. arch wiki is amazing btw
If they were opensourcing it on their own public git repo, we could totally merge request some optimisation from some wizard that give a 10k time increase in a specific library use or something.
the wizard being the programmer, had to mention it.
nvidia is seeing that if they play their cards right, amd may be kicked entirely out of the graphics card space
or possibly they see windows tanking eventually given the gross mismanagement at microsoft
the right things are happening but i doubt they're for the right reasons
Why would AMD be "kicked out of the graphics card space"?
I have two 10-series cards, this is both cool and disappointing. But I understand why I guess.
I truthfully wish Nvidia would go full open source and the code being mostly C or C++, put into Mesa. I've been trying and trying to compile NVK 32-bit on my 64-bit LFS system to no avail as it's written in Rust which heavily complicates things with multiple failure points. I have to be upfront in my book saying that compiling lib32-NVK is a useless endeavor on LFS. Nvidia going open source code really change this. Otherwise without the help of another distro that does things in a weird way like Arch and Gentoo (yes it's all strange when examining it all), then the only real option is the proprietary NVIDIA driver which sucks. And no not using LFS is not a great solution. Everything else works. Shouldn't quit because of one chokepoint. Hopefully things get better from here. This might seem like a hot take.
I just chose AMD over Nvidia because of how annoying Nvidia is with Linux 😢
once i said to Jensen: "make thing work" and he opened up vscode and started to cook right in front of me on his apple watch
you all are welcome and i may continue buying nvidia not because i'm a 3d artist and i'm forced to, but because it will make the lifes of us all nvidia users better in the long term
Hopefully, I will be able to fully switch soon.
I run CachyOS and the devs of Cachy decided the 555 drivers were usable enough and made them available; I installed cachy about 1.5 weeks ago and the install installed the 555 drivers.
I've been running plasma wayland with my laptop that has an nvidia 4050 mobile GPU absolutely FLAWLESSLY and have since fully made the switch.
@@LovecraftianGodsKiller That is so great to hear!!! I just have to port some of my programs though
@@LovecraftianGodsKiller Do you run it in Hybrid Graphics mode or Nvidia only mode?
I had limited options, so this is helpful.
Honestly, open source drivers probably won't happen. Given how much the drivers are crucial for performance.
Just in time for the big switch. nice.
glad i switch to amd :)
Also Intel's dGPUs are looking good drivers-wise
nvidia is nvidia
I got GTX 1060 6G in my desktop, still a great card.
I have been using Nvidia graphics cards for about 20 years and not once had an issue with the proprietary drivers in Linux. Nouveau was, is, and always will be broken. They were broken in 2008, when I first tried nv, and are broken today. OSS for the sake of OSS is a waste of resources.
Sometimes, it's nice to dream that Nvidia is trying to make setting up their GPUs on Linux easier in preparation for the Year of the Linux Desktop.
Okay, dream over. Get back to work.
I love DRM (serious)
Which DRM? Going to keep you guessing
nVidia has seen the self destruction of Windows by Microsoft, and doing what a business does to stay in business. This also is something that is happening already behind the scenes. Their recent move from a Graphic Card company to an AI server company means they are probably using Linux WAY more than any other OS, so this is inevitably going to trickle back into consumer drivers.
Dumb question, but what motivation does Nvidia have to not open source their drivers? Would going FOSS make it easier for AMD or Intel to compete with them? That seems unlikely to me since they've basically got a monopoly on the hardware needed for AI stuff. Assuming it costs nothing to go open source and it improves Linux support, I don't see any reason why they wouldn't do that
At this point dedicating more employees to drastically improve the quality would be a rounding error for Nvidia, they are doing the minimum. At least things are improving.
These are indeed good news, but they sound too good to be true. I'll believe it when I see it.
When you say "novuh", do you mean Nova or Nouveau? I seriously can't tell, your pronunciation is halfway between them.
Nvidia 555 driver still suck on my GTX 2080Ti + Fedora system. Crashes and corrupted gpu memory all the time. 535 remains the best on my machine which forces me to use X11
Let me guess: Falling off the bus?
What about the hdmi 2.1 thing? Remember AMD can't implement it because it's a closed spec. If nvidia becomes open-source will they also lose hdmi 2.1? That'd actually be bad because many gamers need hdmi 2.1.
Best case scenario they'll open source it out of pressure from nvidia.
AFAIK this is a non issue since NVIDIA deals with it in the firmware.
Why not DP?
@@Tracing0029 No TVs have DP.
@felixfourcolor ok, but if it's primarily needed by gamers that is a non issue right since next to no one plays games on their pc hooked up to a TV, and all monitors high spec enough where hdmi 2.1 makes a difference will also have displayport (or usb with displayport alt mode)
So, basically a technology company has to reach the most-valuable status to start looking towards upstreaming... Damn...
Is this a good idea ? Yes,
Well I prefer less rust on my systems as it decays badly and causes automatically terrible stuff.
are you going to talk about systemd removing /home in 256 now? systemd is a joke lmao
> it's real
holy hell poetteringware is a curse on linux
👀
Holy cow... what the actual fuck is this? If systemd declares war on my sweet home, I declare war on systemd.
@@araarathisyomama787 systemd-tmpfiles manages more that just temporary files so if you tell it to purge them all, at least on Debian it will delete your ~
All software has bugs
The "Just Buy AMD" sign off gave me a laugh.
Finally?
Next month will mark 4 years of using Linux on Nvidia without issues. Grub causes more issues than Nvidia drivers, imo. Someday there will be a Zluda or Rcom that can compete with Cuda, but until then Nvidia will be the overpriced king.
cant wait for nv to demand a $30 monthly subscription to unlock specific functions from $1,000+ GPU you paid for.
This is already a thing in enterprise lol
Take a look at NVIDIA GRID prices
@@tablettablete186 big yikes!
i did hear something along those lines.
they are pushing you will own nothing and be happy.
For some reason I thought you were telling Nvidia to buy AMD lol
Can't they just have A.I. maintain the drivers? Wouldn't that be a huge advertisement for them? their A.I. systems write the code.
AI is far from getting to that point.
AI isn't that good.
you want AI hallucinations in your code? lol.
One day, not today lol
Ai isn’t that good.
nVidia is getting there... slowly but surely. And I'd buy AMD, except AMD is always behind nVidia. I'm not going to purchase an inferior product just on principle, I need that performance.
WOOOOOOOOOOO.
the day ngredda is going full open source. is the day. linux is bigger the windows. never. lol
Linux is already used far more than windows, just not on desktops
@@aratof18 yes complete true.
Babe, wake up. A new video drom Brodie just dropped
I hope you wake up once a day
I miss when I could be optimistic about things. Before it became evident that everything any company does will eventually end up being used against us.
I think I'll just go buy AMD
You don´t have to call Nvidia 10 series, "really old gpus." I bought my 1030 new in 2020.
That released 7 years ago lol, in the world of tech that is really old
The date you buy something has nothing to do with the product's actual age.
In the future: "When you want to keep using older hardware Linux won't run anymore, try installing Windows"
Did you just repeat yourself? 😹
Honestly I cannot "just buy AMD" as ROCm is very unsupported and unstable.
colon dee
" just buy AMD" Is kind of a b******* thing to say to the people who require Nvidia technologies but want to use Linux. And "just by AMD" is why Nvidia isn't putting as much effort into this as they should be
I guess what I'm saying is there's a reason Nvidia is a trillion dollar company and AMD isn't despite also being in the CPU market.
Nobody requires Nvidia technologies. The bubble is gonna pop.
Calm down and... just buy AMD. Does wonders for your overall happiness with life as a Linux user.
@@hegemaniac4652 If you would have actually read my comment, you would have understood that there is a lot of people, myself included who can't just buy AMD because they use Nvidia specific technology nearly daily. [Quite frankly, the lack of drivers for Nvidia graphics cards is one of their only reasons. I continue to use Windows at this point]
It's a joke lol, NVIDIA is the most valuable company in the world not because they're a GPU company, but because they're a GPU company in the middle of AI bubble and AMD completely dropped the ball with compute
@@sharp14xYou're right, who even uses CUDA in this day and age!? /j
"just buy amd" ❤
No! Thanks AMD for $3000 "press papier", a useless notebook after your faulty Radeon chips.
noovo
Here i have a hot take: since they are selling so many chips with the ai thing, and open ai server rooms are just gpu, maaaaybe it's easier to mass deploy modules, instead of compiling their own kernel.
1. Modules still need to be loaded by kernel
2. You can compile only in-tree kernel module without compiling rest of kernel
Just buy an AMD 🤣🤣🤣
rust based driver, no thanks. i would rather rewrite the whole kernel in C++.
if you're not developing the driver, why the fuck do you care that it's written in rust? even if you find the community around it annoying, (i do too) that's of zero consequence to you as a user
@@somenameidk5278 rewriting everything that is not C++ on my system is on my to do list. I use gentoo.
@@somenameidk5278 I guess i will rewrite my youtube deleted comment. I use gentoo and on my to do list is to rewrite everything that is not in C++.
Why do you hate Rust?