Nvidia Transitioning To Official, Open-Source Linux GPU Kernel Driver (phoronix.com) 102
Nvidia is publishing their Linux GPU kernel modules as open-source and will be maintaining it moving forward. Phoronix's Michael Larabel reports: To much excitement and a sign of the times, the embargo has just expired on this super-exciting milestone that many of us have been hoping to see for many years. Over the past two decades NVIDIA has offered great Linux driver support with their proprietary driver stack, but with the success of AMD's open-source driver effort going on for more than a decade, many have been calling for NVIDIA to open up their drivers. Their user-space software is remaining closed-source but as of today they have formally opened up their Linux GPU kernel modules and will be maintaining it moving forward. [...] This isn't limited to just Tegra or so but spans not only their desktop graphics but is already production-ready for data center GPU usage.
Demand free drivers! (Score:1)
So it's still non-free software. RMS wouldn't use it, and neither will I.
Be happy for progress (Score:5, Interesting)
Honestly, I never thought we'd see the day when Nvidia open-sourced anything in their drivers. Maybe this means we'll see non-fucked-up support for Optimus in Linux without having to deal with all kinds of stupid shit.
Re:Be happy for progress (Score:4, Informative)
As someone who has been using Optimus on an open-source AMD chip, I can say hands down NVIDIA's implementation has been better.
If anything, perhaps this will spur the open source AMD driver to not suck at it.
The problem, ultimately, is that "Optimus" is complicated.
There are hardware switches, and then there are software sink/sourcing through the iGPU.
The former is easy for everyone. The latter is a bit finicky, and requires environment variables to be set before running something on the dGPU.
On an NVIDIA system, you just setup the binary you want to run in the control panel for the app, and you're done.
I get the impression you're not actually a regular driver of a system with a dGPU and an iGPU. I am. Both AMD and NVIDIA.
Re: (Score:3)
>On an NVIDIA system, you just setup the binary you want to run in the control panel for the app, and you're done.
yes, you are done, with locks most of the time!
i had some kernel+driver versions that worked perfectly, change ANYTHING and you have nothing... i ended playing with the intel card more times than with the nvidia one.
This is the story for the closed source drivers usually, if you get it working, don't change anything... so having a locked distro/kernel setup or pain trying to upgrade to anythi
Re: (Score:2, Troll)
There are shipping laptops right now, with Linux + NV. There are none for AMD.
Re: (Score:3)
There are shipping laptops right now, with Linux + NV. There are none for AMD.
Oh? I'm typing this on an Aspire right now. All AMD, with Radeon graphics, running Debian. There are /lots/ with Linux + AMD shipping now.
Re: (Score:3)
I didn't say you couldn't put linux on such a laptop.
Re: (Score:2)
Pangolin from system76 is one such. It's not from Acer, but then you didn't seem to care in your original post. There: you didn't have long to wait.
Re: (Score:2)
And how the fuck do you figure I don't have long to wait?
I made an assertion, you attempted to disprove it with a falsehood, and then you pointed me at something else that didn't even fit either assertion I made.
Now I will grant you that this far into the thread, the context wasn't 100% clear, but the discussion was about Optimus- i.e., iGPU + dGPU support, and how it difference in Linux between AMD OSS and NV closed source (this is to say nothing about AMD c
Re: (Score:1)
There are shipping laptops right now, with Linux + NV. There are none for AMD.
Oh? I'm typing this on an Aspire right now. All AMD, with Radeon graphics, running Debian. There are /lots/ with Linux + AMD shipping now.
Perhaps you missed it but the discussion here is about "Optimus". Optimus is an NV technology where you have an integrated GPU and a discrete GPU which can be switched between for power efficiency or computing performance. There is no Linux + AMD laptop shipping with such a feature. I understand that, if you completely ignore the context of the discussion, the statement that there are no laptops shipping with Linux + AMD is incorrect but it's a little difficult to participate in a discussion if you ignore t
Re: (Score:2)
again, you also ignore that makes little sense to have a "optimus" setup with amd, you can use directly a AMD APU and have a good performance in a good cpu in a way less complex and cheaper laptop design
Re: (Score:2)
You're just throwing spaghetti at the wall, now.
AMD calls their "Optimus" solution Hybrid Graphics, and yes, they shell tons of them. Just no Linux-focused laptops do so. Being one who, like I said, uses the OSS DRI_PRIME implementation of Hybrid Graphics with a radeon on one of their laptops, I can tell why. This isn't AMDs fault. You can (with much trouble) install Catalyst drivers still, and they work similar to NV. The problem is the OSS stack sucks ass.
No AMD APU is a replacement for ev
Re: (Score:2)
i have used nvidia dGPU and amd APU ... they performance isn't really different ... maybe on high end laptop that weight 5kg you do have a performance difference, but sub 1k laptop (thinks that real people buy), the performance is mostly the same
Starting a game with a environment set isn't hard, hey, it is even easier if you start steam already with it, so any game already used the amd card... but again those setups are rare and stupid today... in the past, AMD cpu were not in the same level as intel, but t
Re: (Score:2)
i have used nvidia dGPU and amd APU ... they performance isn't really different
You're a liar. A flat out liar.
Re: (Score:2)
i have used nvidia dGPU and amd APU ... they performance isn't really different ...
If that was genuinely your experience then you were unquestionably doing something wrong. What exactly were you doing where you didn't notice a difference? I mean if you're just doing email and web browsing then no you wouldn't notice a difference.
maybe on high end laptop that weight 5kg you do have a performance difference
Maybe 15 years ago your assertion was true but I've got a Razer Blade 15 with an RTX 3080 and it weighs just 2kg.
but sub 1k laptop (thinks that real people buy), the performance is mostly the same
Most people buying a sub-$1k laptop aren't interested in GPU performance so getting one with a discrete GPU is pointless when you can just use the inte
Re: (Score:2)
i have used nvidia dGPU and amd APU ... they performance isn't really different ... maybe on high end laptop that weight 5kg you do have a performance difference, but sub 1k laptop (thinks that real people buy), the performance is mostly the same
This is flatly false. The fastest RDNA2 APU there is (6900HS) is slower than the NVIDIA 1650 Max-Q, the a bottom barrel dGPU.
This isn't high end. This isn't a diss on the APUs- for integrated graphics, they're fucking phenomenal, but at the end of the day, they're simply lower power than even low-mid-end dGPUs.
For sub-1K laptops, you generally don't get a dGPU at all, so I'm not even sure how that's relevant.
The discussion was whether or not the APU can compete with dGPUs. It cannot. Not even close. Does
Re: (Score:2)
that is my experience... but no, i do not have a +1k laptop, no idea about those +2k or +4k laptops ... but most people don't use those anyway
Re: (Score:2)
to use a dGPU in a laptop for mail and browser is stupid, of course i'm talking about games, from simpler ones, like project zomboid or rimworld to middle ones, like portal, CSGO, to heavier ones like war thunder, metro redux, Total War games
> Maybe 15 years ago your assertion was true but I've got a Razer Blade 15 with an RTX 3080 and it weighs just 2kg. ... and that without taxes, if you have to pay them... how many people can buy that?
right, a 3.7k laptop
and for 3.7k, you would not use a APU, but also
Re: (Score:2)
right, a 3.7k laptop ... and that without taxes, if you have to pay them... how many people can buy that?
Well where are you setting the goalposts? First it was dGPU, then you moved them so say it was dGPU system less than 5kg, now you have moved them again to say it's dGPU system with a price arbitrarily lower than $3.7k.
and for 3.7k, you would not use a APU, but also top AMD CPU/GPU
No that is wrong, this is the whole point of "Optimus" so that when you're not doing GPU-intensive things it uses the low-power GPU for efficiency.
Re: (Score:2)
a good GPU can power down to low levels. yes, will use more power than a iGPU, but a extra dGPU will also cost more, heavier and something don't really shutdown the dGPU (like those lenovo series, where a termal camera showed that the nvidia dGPU was still hot when "powered off", so consuming energy
Re: (Score:2)
a good GPU can power down to low levels.
Which are you talking about?
yes, will use more power than a iGPU, but a extra dGPU will also cost more, heavier
And? Yes it has more performance so it uses more power, costs more and of course it has at least some mass so yes it will be heavier. What's the problem?
and something don't really shutdown the dGPU (like those lenovo series, where a termal camera showed that the nvidia dGPU was still hot when "powered off", so consuming energy
Then they didn't shut it down properly. If I work purely off the iGPU I get a lot more battery life because the dGPU is switched off, it's not magic.
Re: (Score:2)
Most people do not have problems with nvidia, because they install the driver via the distro and do not change anything until next distro release
Also, linux laptops with AMD, almost anyone can install, it is really plug and play... now with nvidia, if the distro do not help a lot, most people will fail to setup it (or even understand that they need to install the closed drivers!)... so selling linux laptops with nvidia already working is a good selling feature.
Finally, you have Intel laptop with intel integ
Re: (Score:2)
Most people do not have problems with nvidia, because they install the driver via the distro and do not change anything until next distro release
You're 100% correct.
Also, linux laptops with AMD, almost anyone can install, it is really plug and play... now with nvidia, if the distro do not help a lot, most people will fail to setup it (or even understand that they need to install the closed drivers!)... so selling linux laptops with nvidia already working is a good selling feature.
This is correct. Difficulty isn't in installing it, difficulty is with how shitty the stack is for iGPU+dGPU pairing via sink/source. DRI_PRIME sucks.
Finally, you have Intel laptop with intel integrated GPU, AMD laptops with APU, you have intel and amd laptops with nvidia ... intel cpu and amd GPU is possible, but a waste, as you can get the same performance with the APU, but in a much cheaper and less complex setup
This is just ridiculous.
AMD dGPU solutions are far more powerful than their APU solutions. An APU is not a replacement for a dGPU. It's an alternative where you don't need as much power, and a good one. If you're not looking to pay more power for a dGPU, and AMD CPU is a no-brainer in terms of iGPU performance.
having said that, i had a old lenovo laptop that had intel and amd card and dGPU and iGPU worked fine in all kernel and mesa versions... but both cards were weak (15fps in war thunder in the intel, about 22 fps in the amd , this on the time war thunder version, still using the old opengl4 version, much slower than today vulkan engine )
Hah. I bet it was fun buildin
Re: (Score:2)
dGPU to be faster than a APU needs to be a BIG GPU, so top end laptop that is not really a laptop, it a 4Kg brick that you can carry, place in a fan cooled stand and plug to a wall plug... that isn't really a laptop, that is a travel friendly desktop... and usually is very expensive for most people
So just like the high end GPU cards, good to show in expo stands and to show to reporters and youtubers, but either you can find them, or you can pay them... paper releases are easy to do, but that is not what re
Re: (Score:2)
Your argument now seems to be "i don't like dGPUs, therefore they don't exist."
Re: (Score:2)
no, i like dGPU, but most usefull dGPU are way too expensive and heavy laptops... cheaper ones aren't that different from APU
Re: (Score:2)
Most people do not have problems with nvidia, because they install the driver via the distro and do not change anything until next distro release
If you use Ubuntu's software updater it all works just fine, installs new kernel and then recompiles the kernel module automatically via DKMS.
Re: (Score:2)
I've been using some flavor of Ubuntu every day since until earlier this year when I switched to a Mac for my work computer.
Re: (Score:2)
yes, static distros... now go install the latest linux kernel, not the one ubuntu releases... sometimes you need to debug, build something new, a new feature or solve a bug... but ... ohh sorry, that nvidia kernel blog do not compile or crashes...
just because you do not mess with your distro, doesn't mean that nobody do it... and messing with nvidia is a pain, messing with now open source drivers in amd is perfect, you even forgot you are using a advanced graphic card in a linux system
Re: (Score:2)
yes, static distros...
What is a "static distro"? There's nothing "static" about the distro I use, it gets updated kernels all the time.
now go install the latest linux kernel, not the one ubuntu releases...sometimes you need to debug, build something new, a new feature or solve a bug...
I did, because I had a touchpad issue due to a kernel bug. Go learn about DKMS.
but ... ohh sorry, that nvidia kernel blog do not compile or crashes...
What's a "kernel blog"? The nvidia kernel module does compile and doesn't crash, what compilation errors have you been having with the kernel module?
just because you do not mess with your distro, doesn't mean that nobody do it...
Actually I do, it sounds like you're the one who doesn't. But certainly if you do and you have actually had these real problems then post the specific errors.
Re: (Score:2)
>What is a "static distro"? There's nothing "static" about the distro I use, it gets updated kernels all the time.
static as you keep the same versions of software and kernel until the next distro release... as opposing to rolling distros
many rolling distros do track and patch the nvidia blob to work with the new released kernel... but still, you HAVE to use their kernel, you can't choose your own kernel and trust that everything will work
>What's a "kernel blog"?
s/blog/blob/
Re: (Score:2)
static as you keep the same versions of software and kernel until the next distro release... as opposing to rolling distros
Literally nobody does that, people install updated software and kernel updates all the time.
many rolling distros do track and patch the nvidia blob to work with the new released kernel...
No they don't. Feel free to link to such a patch for the nvidia binary blob though.
>What's a "kernel blog"?
s/blog/blob/
Again, you don't understand. What you said was "that nvidia kernel blog do not compile or crashes...", but you don't compile the "blob", it is pre-compiled, that is the very nature of it being a "blob" otherwise it is source code. Source code is what you feed to the compiler, the compiler outputs a binary (sometimes referred to as a 'b
Re: (Score:2)
>No they don't. Feel free to link to such a patch for the nvidia binary blob though.
https://forums.developer.nvidi... [nvidia.com]
>Again, you don't understand. What you said was "that nvidia kernel blog do not compile or crashes...", but you don't compile the "blob", it is pre-compiled, that is the very nature of it being a "blob" otherwise it is source code. Source code is what you feed to the compiler, the compiler outputs a binary (sometimes referred to as a 'blob').
You hare just playing with words now to look
Re: (Score:2)
sorry, pasted the wrong link, this is the one i wanted:
https://forums.developer.nvidi... [nvidia.com]
Re: (Score:2)
You hare just playing with words now to look smart, while ignoring the problem i'm talking about
No I can't understand what you're talking about because you don't know the right words, the words you are using mean something totally different but now that it's clear what you are talking about you can see below how it is exactly the same with AMD and Nvidia open source kernel modules.
fine, the closed source driver (blob) can't be compiled and patch, the kernel shim is the one that dkms compiles and that breaks when you switch kernels and have to hunt patchs to try to fix them...
But it's open source, just like the Intel and AMD drivers which also break when kernel API changes happen and you have to patch them. Hey look, the exact same patch for the open source radeon driver:
https://github.com/ring0 [github.com]
Re: (Score:2)
>But it's open source, just like the Intel and AMD drivers which also break when kernel API changes happen and you have to patch them.
First, it is only open source licensed NOW... just because before you had the source of a small part (the kernel shim source code), the all drivers was not open source... that shim is just a wrapper around a pre-compiled kernel source but will not turn that open source by magic (nvidia did try, but it was denied). That is why the kernel gets tainted after loading the nvid
Re: (Score:2)
First, it is only open source licensed NOW...
I asked you to link to the code you had to patch and that is what you linked to, it's not my problem that you don't know what you're talking about and link to the wrong thing.
Second, when the kernel change, i do not need to change anything, because intel and amd open source drivers are already in the kernel, the kernel is not released until all the included drivers are in sync with any API change ... so all the work is done behind the scene
So your complaint here is just that the open source component of the kernel driver isn't in the kernel tree?
No, what you don't understand that the kernel part WAS NEVER opensource...
But it is open source and when I asked you to link to the problem that's what you linked to. Did you link to the wrong thing?
Mess with anything that the pre-compiled junk needs and you have a broken system
Provide example. Because clearly you have zero experience with GPU computing in the datacenter where it
Re: (Score:2)
>So your complaint here is just that the open source component of the kernel driver isn't in the kernel tree?
ohh yes, my complain is that nvidia don't have a open source driver in the kernel nor help the nouveau one to reach the point where i could be usable!!
While open this kernel kernel driver is a step, is far away to reach that point, will take many years even before nvidia propose any kind of merge with the linux kernel... and until them, is always a pain, where in intel and amd cards, it is always
Re: (Score:2)
nvidia don't have a open source driver in the kernel nor help the nouveau one to reach the point where i could be usable!!
It's open source, they just released it, that's the point of this story that the kernel-mode driver (which is separate from the user-mode driver) is open source. Here is the link to the kernel driver source code https://github.com/NVIDIA/open-gpu-kernel-modules [github.com]
and until them, is always a pain
Again, cite an example. The only example you have provided was a patch to the open source kernel that has always been open source and is the same thing that needed to be patched in the open source radeon driver.
you are still bind that what you compile is not the kernel driver
Ok you are EXTREMELY confused now, let'
Re: (Score:2)
This is the story for the closed source drivers usually, if you get it working, don't change anything... so having a locked distro/kernel setup or pain trying to upgrade to anything that works
Dude DKMS has been a thing for ages, it manages the re-compilation of the kernel module against the running kernel just fine.
Re: (Score:2)
thanks for that info... now try to use that compiled module without crashing if you switch kernels ... sometimes it crashes, others do not even compile... then you have to find online patches to try to solve thoses issues until a official new nvidia drive release is out... but then you also have new kernels, user space, etc
As i said, nvidia clouse source is "good enough" if you are in a static distro and don't change anything... if not, it is usually a pile of trash
With open drivers, i can do whatever i wan
Re: (Score:2)
thanks for that info... now try to use that compiled module without crashing if you switch kernels ...
Ok you need to learn a little more about this, you don't use the same compiled loadable kernel module with a different kernel, you recompile the kernel module against the updated running kernel.
sometimes it crashes, others do not even compile...
Hang on you're not making any sense, you were just saying you were trying to run the same compiled module - so you're not doing it right anyway - and now you're saying that it "do not even compile". So are you saying you are trying to use the same compiled module or that you are trying to compile the module against t
Re: (Score:2)
dkms will build the module for my kernel... but the problem is that the closed driver will not compile in all kernel or sometimes do compile, but crashes.
Just try your self, grab the latest linus kernel, compile it and try to use nvidia with it... good luck!
again, all works fine if you use the main distros and keep their default kernel... but if you start to manage your own kernel (like linus upstream kernel), your nvidia experience will be mostly pain
Re: (Score:2)
dkms will build the module for my kernel... but the problem is that the closed driver will not compile in all kernel or sometimes do compile, but crashes.
You don't compile the closed driver, that's why it's closed rather than open. You don't have the source code for the closed driver so you can't compile it. If you are trying to feed the closed binary to the compiler then I'm not surprised you're having problems.
Just try your self, grab the latest linus kernel, compile it and try to use nvidia with it... good luck!
I literally did that 2 days ago to pull in a fix for the touchpad on my laptop. Built the kernel, built the nvidia kernel module against those updated headers, that kernel module loads the user-mode driver and everything works just fine. The problem
Re: (Score:3)
"The problem" has be been hat NVidia cheats with its drivers. They've repeatedly been caught playing optimizaiton games to enhance their scores with various published tests. Making your drivers entirely open source exposes these cheats. Their cheats have repeatedly been exposed, for decades. If they're moving away frm this practice, good.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
"The problem" has be been hat NVidia cheats with its drivers. They've repeatedly been caught playing optimizaiton games to enhance their scores with various published tests.
AMD cheated on benchmarks by compromising texture quality, and before that they cheated on benchmarks by optimizing specifically for Quake 3 (but no other Q3 engine games). They're both cheaters.
Re: (Score:2)
Er, I should say, before that ATI cheated with the Q3 thing.
Re: (Score:3)
The fallacy you're engaging is sometimes called "whataboutism". It seeks to discard concerns about one abuse or a pattern of abuse by pointing out abuses by another party, often though not always unrelated abuses.
In this case, NVidia may or may not be more guilty of cheating than AMD. But publishing drivers as open source can make cheating, or previous cheating, more apparent, and provides motivation to keep the drivers closed..
Re: (Score:2)
It really doesn't because nvidia has been caught cheating before and will probably be caught cheating again. Not opening the drivers hasn't done that.
Re: (Score:2)
userspace programs can be substituted.
the kernel space is the dangerzone.
Re: (Score:3)
Now now, not all the closed source software is in user space. Some of it is in the GSP too.
They moved as much of the code out of kernel space as they could and then finally opened up the tiny little bit remaining to get rid of the kernel tainting.
Re: Demand free drivers! (Score:2)
And much worse: there is still a huge 34MiB firmware blob that is closed source, encrypted and signed.
Re: (Score:1)
Re: (Score:2)
There's a big difference between firmware and drivers, the drivers run on the host and the firmware runs on the peripheral.
Some devices have both closed, some have open drivers and closed firmware, a precious few have both open. Each has its own implications. If the driver is closed and in the kernel, it could do anything. If the driver is open but the firmware is closed, then the driver is at least auditable, and without its cooperation the device can only do what the IOMMU permits it to do.
In this case on
Re: (Score:1)
Re: (Score:1)
Re: (Score:1)
Re: (Score:2)
Champagne! (Score:5, Funny)
Let's drink to that! Cheers nVidia!
maybe Linus finally got to you!
Re: Champagne! (Score:1)
Hell, Norway froze over...? (Score:5, Informative)
Lets all check the temperatures over there.
It's just the kernel module. The place where the real stuff happens, the xorg drivers, nvidia will probably never open source.
This does make things easier for them though. You still need to check whether the xorg driver supports the xorg or wayland version you're using.
One half truth is one whole lie. Here's the whole truth: without the userspace stuff, the kernel module is still completely useless.
Re: (Score:1)
the user side is the easiest to hammer together with any examples IF you have the kernel side. not many will attempt to make a driver from scratch and to use wayland and xorg either or?
Re: (Score:1)
There's already nvidia support in the user-space, but because you couldn't change clock speeds, using open source drivers were incredibly slow. Open sourcing the kernel side allows changing the clock speeds, so this is a huge deal.
Re:Hell, Norway froze over...? (Score:4)
That is true - the userspace is probably a more critical part of a 3D graphics stack now.
After all, most of the processing heavy lifting happens in the userspace and application side - the kernel driver is pretty much reduced to feeding the beast.
Think about things like Vulkan - compiling shaders and such to run on the chip is a userspace task - your Vulkan application feeds the shader code to the compiler (supplied by GPU vendor), and gets back a compiled binary. It then engages with the kernel driver to load the binary onto the GPU hardware. At which point which side did more work - the kernel loading stuff into memory on the GPU or the userland which took the shader code from the application and compiled it?
Modern GPUs are basically highly micro-programmed parallel processors - there are fewer and fewer fixed-function blocks and more of the graphics pipeline is handled by chunks of code.
It wouldn't surprise me if most of the open-sourced parts was stuff you had to write to manage the card hardware, and very little of it manages the actual graphics pipeline.
In other words, nVidia moved most of the heavy lifting and binary stuff out of the kernel into userspace, letting them get rid of the binary blob that's probably doing very little on modern hardware. The rest is either handled by the firmware on the GPU or binary executables running in userspace.
Re: (Score:1)
Nonsense. As the article points out, nouveau can use it. And you get an untainted kernel.
Re: (Score:2)
lets see, nouveau team is still looking at it and trying to figure what to do with it... because nvidia always "forget" to release proper documentation, so devs have to reverse engineering still... just a little easier than before, but not much
Hacker? (Score:5, Interesting)
Does this have anything to do with the blackmailing hacker who stole all their code?
Re:Hacker? (Score:5, Interesting)
Nvidia has hinted before that the lack of OSS drivers was due to license agreements signed with Microsoft, relating to the original Xbox. Tegra wasn't covered by these agreements, which is how it's had OSS drivers all along.
Re: (Score:2)
Don't trust rumors from a vendor who won't show their source code.
It's in fact a moot point why this is the state of affairs, so I don't know what trust has to do with anything. It's not like Microsoft held a gun to anyone's head and forced them to sign a particular contract... presumably. ;) So they don't get a free pass for their actions, decisions, what have you. This is very low-stakes speculation.
Re: (Score:2)
yes, right... before xbox nvidia already was saying that their closed driver had "license agreements" with other companies and that was why they couldn't... strangely AMD had also those agreements and managed to release a open drivers... they started with a new clean driver and released DOCUMENTATION... they slowly the new open driver linux team worked with the kernel and mesa developers to build the rest.
nvidia could have helped a lot nouveau team by just releasing some documentation and release some signe
Re: (Score:1)
welcome nvidia! (Score:5, Insightful)
Welcome, good news ... FINALLY!! now do that to all your hardware and user space!
This is, IMHO, also a somewhat too little, too late...
You were pushed to it by the DMA-BUF (by the lack of access to, due being closed source and DMA-BUF a GPL-only feature), by losing many big projects due to the lack of open drivers (steam deck, game consoles, big clusters and even datacenters), by looking bad for all these years where AMD and Intel had both open drivers (and even all ARM gpu companies are doing it already) and AMD open driver actually being better than the close source (so nullify the idea that open drivers are always much weaker than close source)... but the lack of proper open user space, still years before being merged in the kernel (if ever, needs open source user space using it to even qualify and pass all code reviews) and being 7 years late is still bad...
So this is the first step of a marathon... a important step, but a long way to go
Re: (Score:3)
ohh, i forgot, also pushed by the lack of proper wayland support, so many problems that ubuntu reverted nvidia to X11 support
Re: (Score:2)
Who cares about Wayland support?
Re: (Score:3)
Yeah that. Running X11 has zero drawbacks for me. I get frame rates about the same on Linux as on Windows for the same games so clearly I'm not leaving any meaningful graphics performance on the table. What would I get out of Wayland? Zilch.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
So this is the first step of a marathon... a important step, but a long way to go
Nope. It is too late. After 20 years of wanting only Nvidia and NOT ATI/AMD, I finally gave up. Too little, too late. Good bye Nvidia. Keep your money, I hope it makes you feel all warm and fuzzy as you sleep. I have no more interest in anything you offer.
Re: (Score:1)
OK, a little late.... (Score:1)
Well it's a partial gotcha. Looks like Kernel mode only, no userspace. (Yet.) Let's hope OSS user space comes soon, or maybe nouveau can provide userspace support?
WTF?! This isn't really new (Score:1)
The nvidia kernel module source has always been available. What were we compiling all these years?
This is just a license change. Big whoop. The important stuff is still closed as before.
Re: (Score:2)
The nvidia kernel module source has always been available. What were we compiling all these years?
You've been compiling boiler plate code over a huge binary kernel module.
Re: (Score:2)
Which won't change much with this OSS release...
Per the referenced article:
> the GSP is binary-only firmware loaded at run-time. The open-source kernel driver explicitly depends upon the GSP-supported graphics processors.
GSP - Gpu System Processor, basically a RISC-V core that manages the GPU.
From what I understood, this release is a 900 function API translator into calls the GSP understands, and I assume that quite a lot of the original binary blob code (or something similar) will still be running in th
Re: (Score:2)
Of course, this reorganization also paves the way for distribution inclusion. linux-firmware package contains lots of unmodified binary redistribution allowed content.
Quality wise it sounds like they aren't ready for that yet, but it is a model consistent with a lot of other devices with proprietary hot-plug firmware.
So while not perfect, it's pretty consistent with most hardware that has proprietary on-device code executing, including hard drives, network chips, even the processors themselves.
Re: (Score:2)
They'd need to opensource at least some userspace code that would use the kernel driver.
Re: (Score:2)
Depends, was unsure exactly what the deal was.
They have their own GL/vulkan libraries. It may be possible for Mesa to target the nVidia drivers. nVidia may provide *better* absolute GL/Vulkan performance, but Mesa could provide a very serviceable experience on par with their support for other platforms, enriched by being able to target nvidia drivers and this GSP rather than the best nouveau can come up with. This means a future 'out of the box' experience that's quite compelling, with high probability of
Re: (Score:2)
> They have their own GL/vulkan libraries.
Yes, closed source.
AFAIK the deal is that in mainline everything must have an OSS userspace user.
I understand they're basically hoping to create an amdgpu (pro) scenario for themselves. It took AMD 10 years. And currently the opensource stack is actually better. (except for compute, which for some reason no one in FLOSS is interested in)
Yes, the out of the box experience will be better, on paper its a positive move, but my compute freedom loving soul cries when I
too little, too late (Score:4, Informative)
Can't say I care all that much. Even aside from the fact that the userland portion of the driver (i.e. most of it) is still closed-source proprietary binary blob, this new kernel module requires Turing (e.g. RTX 20xx series cards) or newer.
My current nvidia video cards are a 1060, a 970, and couple of 750s (Turing was too expensive to bother with, and Ampere was even worse). They won't be supported, and I've already decided that unless there's some major fuckup with the pricing or availability of next-gen cards, my next GPU upgrade will be to RDNA3's low-end Navi 33 - a 7600 or 7700 or whatever they end up being called.
Given that a low-end navi 33 is supposed to be roughly comparable in performance to a current high-end 6900 XT or RTX 3090, that will be a massive performance boost compared to my current 1060 6GB. That will be overpowered for what I currently need, as I have a 1440p 60Hz monitor and won't be upgrading to a 4K monitor for quite some time.
Re: too little, too late (Score:1)
Re: (Score:1)
Re: (Score:1)
Re: (Score:1)
It'll happen the day you figure out how allow copyrighted and trademarked code that they license from third-parties be GPL compatible.
Re: (Score:2)
Can't say I care all that much. Even aside from the fact that the userland portion of the driver (i.e. most of it) is still closed-source proprietary binary blob
More critically the userland portion of the driver is where 95% of everything happens. Every time a new game is released NVIDIA don't push out a new game ready driver because something fundamentally changed the way the kernel needs to talk to the hardware. No it all happens a level above.
An open source driver in this case is kind of like running Windows inside a VM on a Linux kernel and talking about how OSS friendly you are.