NVIDIA Pushes 62MB Of GSP Binary Firmware Blobs Into Linux-Firmware.Git

[Rovano]

Rovano it is worse this time. AMD is selling brand new laptops with Vega integrated graphics not 2 years ago, not 1 year ago, but *this year*. Go check out Ryzen 7 7730U. It is a 2023 product.

I understand it. You could tell me straight that it was an architecture of 2017. So it's insolent from the manufacturer.

[qarium]

The negativity may be irrelevant to Linux users. But the retirement is causing real annoyance to Windows users.

you know that the windows users get what they deserve... we all know that microsoft is a hostil enemy company and those people who use to use microsoft windows deserve all any every pain they get.

but there is a movement wo not use AMD drivers for AMD hardware remember the day of ATI Omega drivers ?

see here for the Omega Reborn drivers.

https://www.amernimezone.com/

in the near future with more and more parts opensource for the driver this kind of drivers could become more and more advance.

about my first sentence we all know that microsoft is hostil enemy company and anyone who use windows deserve the pain they get.

[billyswong]

I understand it. You could tell me straight that it was an architecture of 2017. So it's insolent from the manufacturer.

Well, turns out they put a "Zen 3" into the CPU part of 7730U, so not exactly 2017. And upon more inspection, they let 7730U share the same driver package of the newer processor series, and offer the legacy driver package when one look for drivers for 5xxx series or before. So they are somewhat "selective" when they cut off support of Vega.

[mdedetrich]

https://answers.microsoft.com/en-us/windows/forum/all/nvidia-drivers-not-working-on-windows-81-help/412cc921-b9c2-4c34-9f57-263d4ef6f882

Except what you just wrote if you go looking you find that Nvidia had issues at Windows 8 and 8.1 as well. Microsoft implements the final solution in Windows 10 where to sign drivers Nvidia and any other vendor has to be update on driver ABI support.

There has been nothing serous or wide spread with Windows since Windows 10 when Microsoft implemented to sign drivers you had to log into a Microsoft portal and do it there and that portal refuses to sign drivers that are not up to date on ABI support.

Cherry picking random isolated issues with NVidia doesn't prove your point, you should stop doing it. Its not hard to google "AMD driver issue due to <X>" and find something to "support" my argument.

@avix Is right here, you are really scraping the barrel hard when it comes to find whatever thing possible to try and sling something on NVidia

[mdedetrich]

The only reason most AMD users are not using AMDPro is because it's not been hassle free at all and then AFAIK (I can be wrong, I don't remember it well now) it had its own kernel module with which people struggled with and AMD did a very poor job maintaining it.

NVidia on Linux also has its own kernel module in the blob, the difference here is that NVidia puts all of its engineering resources into supporting that blob (including the module) to be as painless /easy to use as possible on Linux where as AMDPro is a relatively recent thing for AMD and due to the OS driver being in tree, the only people that use it are some professional users (also same reason why AMDPro is only supported on some ditstros)

[oiaohm]

NVidia on Linux also has its own kernel module in the blob, the difference here is that NVidia puts all of its engineering resources into supporting that blob (including the module) to be as painless /easy to use as possible on Linux where as AMDPro is a relatively recent thing for AMD and due to the OS driver being in tree, the only people that use it are some professional users (also same reason why AMDPro is only supported on some ditstros)

The AMDpro installer including kernel module is not new.

https://www.amd.com/en/support/kb/release-notes/rn-radpro-lin-16-40

AMD has been doing this ever since they discontinued the ATI fglrx drivers.

https://www.phoronix.com/forums/forum/linux-graphics-x-org-drivers/open-source-nvidia-linux-nouveau/1421074-nvidia-pushes-62mb-of-gsp-binary-firmware-blobs-into-linux-firmware-git?p=1421437#post1421437

Yes AMD bridgman was very clear this was the case.

https://forums.developer.nvidia.com/t/stop-asking-simple-users-to-install-the-unfriendly-run-file/48586

Now read the above Nvidia post.

There is a repeating problem. People are told to use the Nvidia driver versions that their distribution provides. Linux distributions custom patch their kernels so breaking compatibility with NVIDIA and AMDpro direct provided drivers alot. AMDpro latest userspace does work with quite a few versions older or newer kernel driver with Linux. Nvidia does not this does make Nvidia user-mode drivers more limited when you run into kernel issues.

Due to AMD amdgpu and radiion drivers providing a stable ABI to userspace under Linux unless something wrong with the kernel driver you are not going to install the AMDPro included kernel modules instead just install the AMDPro userspace.

The reality here is the AMD stable interface between kernel space and userspace with their kernel drivers for graphics makes updating the kernel mode driver less important. So lots of distributions only have the instructions to install the AMDpro userspace and skip the AMDPro included kernel mode drivers.

mdedetrich reality majority of the time with a Distribution with upto date kernel version you have a newer version of the AMD kernel module driver than what was include with the AMDpro driver installer. As bridgman note only enterprise distributions and the like staying on older kernels in most cases installing AMDpro include kernel mode driver make any sense. Everything else the kernel included driver is newer. Remember AMD driver most current version backported to all Linux kernel LTS version updates. That does not leave many distributions users should be still using using old enough amd kernel driver that need to be replaced by the AMDpro include amd kernel driver.

[avis]

NVidia on Linux also has its own kernel module in the blob, the difference here is that NVidia puts all of its engineering resources into supporting that blob (including the module) to be as painless /easy to use as possible on Linux where as AMDPro is a relatively recent thing for AMD and due to the OS driver being in tree, the only people that use it are some professional users (also same reason why AMDPro is only supported on some ditstros)

Thank you. I thought I was going mad.

There is a repeating problem. People are told to use the Nvidia driver versions that their distribution provides. Linux distributions custom patch their kernels so breaking compatibility with NVIDIA and AMDpro direct provided drivers alot. AMDpro latest userspace does work with quite a few versions older or newer kernel driver with Linux. Nvidia does not this does make Nvidia user-mode drivers more limited when you run into kernel issues.

There's no problem. I guess fewer than 1% of Linux users install NVIDIA drivers using the provided SH file and then as I've been an NVIDIA user for over two decades, I can count on my fingers the number of times NVIDIA dragged their feet supporting the latest mainline version of the kernel which is really nothing which is actually an amazing job.

Imagine Microsoft breaking kernel APIs/ABIs every 3 months. The whole world would go crazy.

[oiaohm]

There's no problem. I guess fewer than 1% of Linux users install NVIDIA drivers using the provided SH file and then as I've been an NVIDIA user for over two decades, I can count on my fingers the number of times NVIDIA dragged their feet supporting the latest mainline version of the kernel which is really nothing which is actually an amazing job.

AMD open source kernel drivers do that all the time as well. AMD manages to support the Linus branch because they are mainline. Nvidia does not in fact.

Imagine Microsoft breaking kernel APIs/ABIs every 3 months. The whole world would go crazy.

Yet Nvidia doing the same thing with every binary driver release is perfectly fine.

Avis there is a few things to remember.
With Windows Vista microsoft wanted opengl and any future item to use direct X kernel to userspace ABI to make this interface stable. Microsoft work on dzn and d3d12 in mesa3d at the moment is moving in that direction.

Linux kernel rule do not break userspace. Yes all mainline kernel drivers have to play by this rule. This means no version locking between your userspace and kernel driver where possible. So what Nvidia been doing with their drivers is counter to what Microsoft kernel developers and Linux kernel developers want. Yes both Microsoft and Linux kernel developers agree there should be a stable ABI between kernel space and user-space.

Also avis you know the fewer than 1 percent

Lets point out the elephant in the room.

https://packages.debian.org/sid/nvidia-driver 525.125.06-3 seams current right.
https://www.nvidia.com/en-us/drivers/unix/ but then 535.129.03

Yes avis like it or not there is a problem. The reality is the

24 July 2023 is the release date of 525.125.06 and the current is 2 Nov 2023 535.129.03.

The majority of Nvidia users on Linux don't use the latest version of Nvidia graphics drivers. Nothing to be 3 to 6 months out of date.

If you are measuring age of graphics driver users are using AMD open source mainline driver gets to majority AMD Linux users faster than the Nvidia closed source does for Nvidia closed source users.

Yes you want to talk about Nvidia supporting the latest kernel the reality is Nvidia is not being able to get there driver to end users effectively on Linux because the latest distribution kernel has patches and just supporting mainline with closed source driver does not mean it works.

AMD open source mainlined drivers is more effective getting the newest version of the kernel driver to users faster. The fact linux users normally has newer version of the kernel driver than what can be bundled with the closed source userspace with AMD results in AMD invested less time into packaging up kernel drivers on Linux with the closed source userspace because what is the point.

avis think about it what you have called amazing job is lots of hours of investment only to get out of date versions to users more often than not.

Nvidia is not looking at open sourcing their kernel mode driver for no good reason. Just because a party invest decades of develop time into something does not mean it the correct way or the most effective way.

[avis]

Yet Nvidia doing the same thing with every binary driver release is perfectly fine.

NVIDIA kernel driver and userspace are one thing. What you're saying is 100% technically incorrect and thus equally wrong. You always masterfully find ways to portray NVIDIA as a bad/incompetent/greedy company - I'll give you that. Too bad out of these three words all three equally apply to Intel and AMD one way or another. Also, what would you say to the fact that NVIDIA drivers have been ranked the best in terms of stability and features for over two decades excluding the transition to Vista?

And I couldn't care about the versions of NVIDIA driver in Debian. That's a Linux/Debian issue, not NVIDIA's. I'll tell you even more: Fedora officially doesn't support/contain/supply the Linux NVIDIA driver. According to you we shall blame NVIDIA for that, shan't we?

NVIDIA doesn't create 1000050000 drivers for 1000050000 Windows versions, they have just one. Yeah, Windows 10 and 11 have largely the same graphics driver API with some added features in W11. The vast majority of W11 drivers work in W10 without any changes.

The lengths at which Linux fans are ready to go just to vindicate the Linux kernel broken development design are always astonishing. And before you say this design is OK, it's NOT, as new kernel releases sometimes bring nasty regressions because drivers haven't been properly updated for new API changes. Of course in a fantasy world the Linux kernel is free from regressions.

I'm not an NVIDIA fanboy but I'll give credit where it's due. You on the hand will never admit anything good about NVIDIA and will always portray AMD/Intel as the best graphics companies. Too bad the world disagrees with you. People don't buy shit even if it has a reputation.

And the rest is just complete whataboutism not pertaining to this discussion in any shape or form.

[oiaohm]

Also, what would you say to the fact that NVIDIA drivers have been ranked the best in terms of stability and features for over two decades excluding a transition to Vista?

Ranked on what avis.

https://linuxgamingcentral.com/posts/moving-from-nvidia-to-amd/

The reality here Nvida has not had stability on Linux not at the same level as AMD.

And I couldn't care about the versions of NVIDIA driver in Debian. That's a Linux/Debian issue, not NVIDIA's. I'll tell you even more: Fedora officially doesn't support/contain/supply the Linux NVIDIA driver. According to you we shall blame NVIDIA for that, shan't we?

This means you just want to stick you head in sand and ignore the user delivery problem. I did not say blame Nvidia. This is more you have to look method AMD and Nvidia is using on Linux and what one is being more successful at getting newest drivers delivered to users. AMD method on Linux like it or not is more successful than the Nvidia one.

NVIDIA doesn't create 1000050000 drivers for 1000050000 Windows versions, they have just one. Yeah, Windows 10 and 11 have largely the same graphics driver API with some added features in W11. The vast majority of W11 drivers work in W10 without any changes.

Again ignoring the delivery problem and the requirements at play. Lets say I wanted to release a new W10 driver today. Yes a pure W10 driver. I log into the Microsoft portal for signing drivers and attempt sign pure Windows 10 driver guess what rejected no sign for you. To sign a driver W11 driver ABI features have to be there is the error message. I want to release a pure W11 driver with no W10 compatibility parts log into the portal and attempt to sign it guess what again driver reject backwards compatible W10 ABI has to be there or you are not signing driver message now. Yes this is case you have to-do what Microsoft wants or you driver is not signed.

Note how you said vast majority of W11 drivers work on W10 not all. By the requirements to get driver signed in theory all W11 drivers should work on W10 but why don't they. That right the Windows 10 kernel can have been updated since the time W11 driver was signed so changing the ABI slightly so breaking the driver. Also you have a Windows 11 driver signed new Windows 11 kernel releases and the driver does not work correctly any more same issue slight ABI changes causing breakage.

The lengths at which Linux fans are ready to go just to vindicate the Linux kernel broken development design are always astonishing. And before you say this design is OK, it's NOT, as new kernel releases sometimes bring nasty regressions because drivers haven't been properly updated for new API changes. Of course in a fantasy world the Linux kernel is free from regressions.

https://answers.microsoft.com/en-us/windows/forum/all/after-recent-windows-update-drivers-break-after-a/55e0e858-b4c2-441f-ac95-2a80973af67b

This is another point you need to pull head out sand. Microsoft releases a new Window kernel. You have old Windows drivers manually installed MS Windows breaks. What is written stable api nonsense document effects Microsoft as well. New kernel releases bring nasty regressions because drivers have not been properly updates is a Windows, Linux, Mac OS ... problem.

Somehow to reduce this problem you have to lock driver version and kernel version with each other. What is written in the Linux kernel called stable-api-nonsense.rst is a stack of technical problems that you run into when you attempt to make a stable kernel ABI/API and use a monolithic kernel base this includes windows nt hybrid design. Microkernels that run there drivers in user-space don't have this problem but there is a performance cost to this. Microsoft Windows is effected by the issues listed in stable api nonsense and it very simple to find examples of proven it.

I'm not an NVIDIA fanboy but I'll give credit where it's due. You on the hand will never admit anything good about NVIDIA and will always portray AMD/Intel as the best graphics companies. Too bad the world disagrees with you. People don't buy shit even if it has a reputation

https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam?platform=linux

No you are a fan boy birdie. You will find any number that saves you ass. When talking about desktop Linux Nvidia DGPU is not the dominate installed. AMD is the dominate DGPU. So what is AMD doing right so they are the dominate Linux desktop DGPU.

Yes we can also list everything AMD drivers do on Windows not to be popular on Windows. Nvidia is not getting their Linux side that right. AMD is not getting their Windows side that right.

Platform integration is important factor for market share. AMD has better platform integration on Linux than Nvidia does and that shows in market share numbers. Nvidia normally has better platform integration on Windows than AMD and this again shows in the market share numbers.