Nvidia... nvidia-drivers... what's having a nvidia gpu like?

[eccerr0r]

I just got my hands on a "new to me" Nvidia GPU, an old P1000 GPU.
This appears to use the 575-series of nvidia-drivers ... but what does this mean?

I had a GTX660 that needed the 470-series nvidia-drivers but it got deprecated by nvidia. This card was not in good shape so no big loss.

This Pascal card should be in decent shape, was running the 575 driver and it works well. What I'm wondering...

How often does nvidia give up on nvidia drivers, meaning switch major versions due to a fundamental shift in their newest GPUs? Should I expect this Pascal GPU be put on the backburner soon? I suppose the GTX660 is old, but just a few years older than the Pascal (though I'm seeing reports that the two GPUs are similar in speed somehow?) Is this board going to be dropped soon too or will there be some life in it as newer boards still have a similar architecture?

Yeah this board is a bit faster than my ATI R7-250. Noticeable though not extremely faster, as hinted by video cards available today that are 10-20x faster than even this Pascal board...

Last time I was using an nvidia card with nvidia-drivers excepting the GTX660 that I never really used as it was acquired defective... was a long long time ago. Almost thought it was my GeForce4 MX420 old, before I ran Gentoo, and it was a major PITA keeping up with kernel versions. Seems the Gentoo devs have gotten it more seamless which is a good sign, but will I be disappointed by nvidia-drivers within the year? 2 years? 5 years?

[Banana]

I was a long time nvidia card user (private and work), but can not really remember the card details.
I had never problems with the drivers (used the official one) and all the time it worked over the timespan of 6-8 years. The work card was also some of the workstation quadro cards.

Your is also a Quadro card and maybe this could lead to a longer support of drivers.
It is even listed on the HPE driver download page
Nvidia itself has also an archive of its drivers.

So, for it looks good and don't think support will drop soon.

[eccerr0r]

Cool, hope this card will last a while. Was a bit concerned because this board is already ~ 8 years old, so need to add that onto the counter. The failure of the GF4MX420 left a bad impression back then coupled with the closed source annoyances, I went ATI ever since despite having to use catalyst (also closed source) but subsequently had a RadeonHD 3650 and 5770 fail on me, so it's not just nvidia (Both the GF4MX420 and GTX660 croaked. If there's any piece of hardware that I have bad luck with it's GPUs, and then hard drives.) Recently I've been getting old/used hardware for cheap so I guess it doesn't hurt as bad when they fail.

[dmpogo]

I am nvidia card and nvidia-driver user since, I think 2003 or so. Never had any issues. Nvidia supports its drivers much longer than Gentoo in protage. You can still download from nivida 71.86.15 drivers, which are like 20 years old.

[Ionen]

I am nvidia card and nvidia-driver user since, I think 2003 or so. Never had any issues. Nvidia supports its drivers much longer than Gentoo in protage. You can still download from nivida 71.86.15 drivers, which are like 20 years old.

I don't think that keeping old drivers downloadable is called support. These are broken with current Xorg (let's not even mention wayland), current kernels, modern compilers, and filled with security vulnerabilities among other issues. Essentially useless unless you're using a linux distribution from 20 years ago as well (which could be Gentoo by using the tree from that time assuming can find all distfiles needed).

NVIDIA also officially calls these unsupported themselves, currently they support (as in, still do new releases) the 535.x branch and newer for production branches (535 is probably about to end though, but that one is not really needed for old hardware unlike 470 is given latest 570 still supports all the same cards).

Edit:
On a side-note, imagine the next time they drop support for cards it'll be everything not using GSP (pre-GTX 1650), not that we won't have a driver branch supported for several years to keep them going for a while -- on the downside, nouveau (or nova) support improvements also require GSP so these older cards won't get good open source support either (but that also means newer ones may eventually get great support).

That aside, nvidia has never really given me trouble either I do replace my PC every 10-15 years or so as needed and don't keep (very) old hardware around much. I'd personally avoid nvidia if I wanted a laptop again though, for laptops I'd be happy as long as the display works and some integrated intel gpu is good enough.

[dmpogo]

Edit:
On a side-note, imagine the next time they drop support for cards it'll be everything not using GSP (pre-GTX 1650), not that we won't have a driver branch supported for several years to keep them going for a while -- on the downside, nouveau (or nova) support improvements also require GSP so these older cards won't get good open source support either.

That aside, nvidia has never really given me trouble either I do replace my PC every 10+ years or so and don't keep (very) old hardware around much. I'd personally avoid nvidia if I wanted a laptop again though, for laptops I'd be happy as long as the display works and some integrated intel gpu is good enough.

I don't have nor plan to have nvidia on laptops. One of my desktops is from 2007, with original NVIDIA Corporation G86 [GeForce 8500 GT] from the same year, using nvidia-drivers-340.108-r101 (from overlay).
Works just fine with the latest Xorg and 5.15 kernel. Limiting problem is that the card can't get out more resolution that Full HD and my displays are better That is almost 20 year old card. Show me better support somewhere.

My other desktop is from 2011, but a newer card, NVIDIA Corporation GK208B [GeForce GT 730] running nvidia-drivers-470.256.02-r2 (still in tree but masked). No problems either.

[eccerr0r]

Agreed, support means they will update the driver to work with modern kernels, Xorg, Wayland, etc., not just having it available and forcing one to use Linux Kernel 1.2.11 ... caveat of being a binary driver, can't do much with binaries when ABI changes.

All I have to say is that nouveau is a stop gap solution to prevent an nvidia card from becoming a paperweight. Seems to drop 1/2 to 2/3 of the frames per second versus nvidia-drivers.

I have an 8200 somewhere but its performance under nouveau is pretty awful. I also have a Quadro NVS295 but I don't think that works under any nvidia-driver either...

Granted I have to say that the R200 Radeon OSS support has been broken and also unsupported which is a shame. They work as a simple 2d accel at least but 3d is broken. At least my junky RadeonHD 6350 and 6450s are supported and work as well as the cards work for what it's worth, and still can run on latest kernels. And faster than the 8200's under nouveau.

I guess I'll try to use the P1000 as much as I can and hope it breaks before nvidia decides to stop supporting it, and save my Radeon R7-250 which hopefully will be supported longer.

[dmpogo]

Agreed, support means they will update the driver to work with modern kernels, Xorg, Wayland, etc., not just having it available and forcing one to use Linux Kernel 1.2.11 ... caveat of being a binary driver, can't do much with binaries when ABI changes.

It is a bit of chicken and egg. One can as well say that it is a modern kernel that drops support for fully functional old hardware with working drivers.

That is why after buying new hardware and achieving a state that everything works, I stop updating kernels, as long as possible. The job of the kernel is to run hardware, and if it can't anymore, I am not chasing it, I am sticking to the one that can.

[eccerr0r]

It's also got the responsibility to keeping software that's not supposed to run from running... hence the reason for updates. The hope is that the computer is no longer useful comes time new kernels will no longer run on it. That said I still have a lot of old machines running recent/latest kernels and had to dump hardware that was no longer supported...

[dmpogo]

I am nvidia card and nvidia-driver user since, I think 2003 or so. Never had any issues. Nvidia supports its drivers much longer than Gentoo in protage. You can still download from nivida 71.86.15 drivers, which are like 20 years old.

I don't think that keeping old drivers downloadable is called support. These are broken with current Xorg (let's not even mention wayland), current kernels, modern compilers, and filled with security vulnerabilities among other issues. Essentially useless unless you're using a linux distribution from 20 years ago as well (which could be Gentoo by using the tree from that time assuming can find all distfiles needed).

NVIDIA also officially calls these unsupported themselves, currently they support (as in, still do new releases) the 535.x branch and newer for production branches (535 is probably about to end though, but that one is not really needed for old hardware unlike 470 is given latest 570 still supports all the same cards).

Edit:
On a side-note, imagine the next time they drop support for cards it'll be everything not using GSP (pre-GTX 1650), not that we won't have a driver branch supported for several years to keep them going for a while -- on the downside, nouveau (or nova) support improvements also require GSP so these older cards won't get good open source support either (but that also means newer ones may eventually get great support).

That aside, nvidia has never really given me trouble either I do replace my PC every 10-15 years or so as needed and don't keep (very) old hardware around much. I'd personally avoid nvidia if I wanted a laptop again though, for laptops I'd be happy as long as the display works and some integrated intel gpu is good enough.

Ok, it came after me - I cannot compile my 340.x driver after upgrade to gcc-14. I believe a year ago you had a discussion of that on nividia forum - it is this -Werror-implicit-function-declaration.
As you were saying there, the only way is to disable it - could you advise me how to do it ?

Thanks !

PS leaving my ancient nvidia aside, I am a bit annoyed were gcc-14 is heading . Now I have portage refusing to update virtualbox kernel module, because my kernel is compiled with gcc-13. It is a first time in 20 years that I see such a conflict. I'd hate to see gcc going rust path where you need to match the versions at every turn.

[flysideways]

I still have the P400 Quadro that was the first gpu in my 12 core Ryzen that I built back in 2020. Partly it was because video cards were so hard to find, but my desk also had three 2k displays on it. It was the first card I'd had with three DP outputs. I no longer had to mix and match HDMI and DP inputs to the monitors.

I just used nouveau. It did everything I wanted.

It eventually got replaced with a RX 5500 XT that also had three DP ports.

Right now the P400 is just collecting dust on a shelf. I liked it and have been slow to get rid of it, intending to repurpose it.

But, a Pi 5 makes a lot of my old stuff look, well, old.

[Hu]

PS leaving my ancient nvidia aside, I am a bit annoyed were gcc-14 is heading . Now I have portage refusing to update virtualbox kernel module, because my kernel is compiled with gcc-13. It is a first time in 20 years that I see such a conflict. I'd hate to see gcc going rust path where you need to match the versions at every turn.

This is not a gcc problem. For as long as I can remember, the Linux kernel has enforced a rule that modules shall only be loaded if their magic string matches the main kernel exactly.

[Edit: the below was originally part of the preceding paragraph, but is incorrect as of commit 1bb4996bcebca1cde49d964b4e012699ce180e61, so it does not apply to any still supported kernel. It is retained for context, since later posts refer to it.

• Part of the magic string is the compiler version, hence you must build the main kernel and the virtualbox modules with the same gcc version. Build the virtualbox modules with gcc-13, rebuild your kernel with gcc-14, or switch to a hypervisor (such as Qemu KVM) that does not need out-of-tree modules.

- End incorrect section.]

[dmpogo]

PS leaving my ancient nvidia aside, I am a bit annoyed were gcc-14 is heading . Now I have portage refusing to update virtualbox kernel module, because my kernel is compiled with gcc-13. It is a first time in 20 years that I see such a conflict. I'd hate to see gcc going rust path where you need to match the versions at every turn.

This is not a gcc problem. For as long as I can remember, the Linux kernel has enforced a rule that modules shall only be loaded if their magic string matches the main kernel exactly. Part of the magic string is the compiler version, hence you must build the main kernel and the virtualbox modules with the same gcc version. Build the virtualbox modules with gcc-13, rebuild your kernel with gcc-14, or switch to a hypervisor (such as Qemu KVM) that does not need out-of-tree modules.

I never seen that before, and I tend to have kernels running for 5 years (previous, still 4.19 was going from 2018 to 2024) with virtualbox and nvidia-drivers updated along the way, pretty sure with newer gcc. But I need to investigate. I have a machine that runs 5.15.158 kernel for the last year

Code: Select all

 cat /proc/version
Linux version 5.15.158-gentoo (root@wheeler) (gcc (Gentoo 13.2.1_p20240210 p14) 13.2.1 20240210, GNU ld (Gentoo 2.41 p5) 2.41.0) #1 SMP Tue May 7 14:20:17 MDT 2024

It also has virtualbox-7.1.4-r1 installed.
Couple of days ago I did routine upgrade, where new version of virtualbox-modules-7.1.8 was in the list. It failed first giving the warning

Code: Select all

 * Warning: kernel 5.15.158-gentoo is built with gcc-13 but
 *          '/usr/bin/x86_64-pc-linux-gnu-gcc' is gcc-14
 * This *could* result in build issues or other incompatibilities.
 * It is recommended to either `make clean` and rebuild the kernel
 * with the current toolchain (for distribution kernels, re-installing
 * will do the same), or set the KERNEL_CC variable to point to the
 * same compiler. Note that when it is available, auto-selection is
 * attempted making the latter rarely needed.

and then falied with

Code: Select all

././include/linux/kconfig.h:5:10: fatal error: generated/autoconf.h: No such file or directory
    5 | #include <generated/autoconf.h>
      |          ^~~~~~~~~~~~~~~~~~~~~~
compilation terminated.

Now that indeed could be not gcc issue, but after this warning I am not sure. Attempt to rebuild already installed virtualbox-modules-7.1.4-r1
gives the same warning and fails with the same errors

Could be that my sources are corrupted somehow, but for one I do not see why they would, all the files and links are from may-summer 2024, so I have not been there for a while

[rab0171610]

viewtopic-t-1083456-start-0.html

[Banana]

viewtopic-t-1083456-start-0.html

A little more context why you are refering this topic is always helpful

[rab0171610]

You are right. It would be helpful.
@dmpogo
To reiterate what Hu mentioned. You need to rebuild your kernel source. What Hu detailed about gcc versions and out of tree kernel modules such as Nvidia drivers or VirtualBox modules has been discussed previously. See the direct link above.

Or else you ran 'make distclean' on your kernel sources in the past inadvertently, which is the reason for the missing /usr/src/linux-*/include/generated referred to in line #5 in the file of kernel sources ././include/linux/kconfig.h which is listed in your error message.
That directory is populated during kernel compilation.

If the /usr/src/linux-*/include/generated directory is present in your kernel sources and populated and autoconf.h exists therein, then it is likely that the virtualbox modules and/or the kernel source need to be rebuilt with the same gcc version.

Regardless, the easiest thing to do is to just rebuild your kernel and modules.

[dmpogo]

You are right. It would be helpful.
@dmpogo
To reiterate what Hu mentioned. You need to rebuild your kernel source. What Hu detailed about gcc versions and out of tree kernel modules such as Nvidia drivers or VirtualBox modules has been discussed previously. See the direct link above.

Or else you ran 'make distclean' on your kernel sources in the past inadvertently, which is the reason for the missing /usr/src/linux-*/include/generated referred to in line #5 in the file of kernel sources ././include/linux/kconfig.h which is listed in your error message.
That directory is populated during kernel compilation.

If the /usr/src/linux-*/include/generated directory is present in your kernel sources and populated and autoconf.h exists therein, then it is likely that the virtualbox modules and/or the kernel source need to be rebuilt with the same gcc version.

Regardless, the easiest thing to do is to just rebuild your kernel and modules.

Sure, though recompiling the kernel is not the easiest option in my case. I can not take this computer down until at least the Fall.

Anyway, that should have been in a separate thread, my fault. My main question was to Ionen, who, I think, knows how to compile old nvidia drivers with gcc-14 with now elevated 'implicit_declaration' to error status.

BTW, discussion at the link above concluded that out of tree modules DO NOT require the same gcc version to be functional as a rule, and gcc version is not a part of magic string.

[eccerr0r]

Well it looks like it happened.

The latest december nvidia-drivers supposedly dropped support of my Quadro P1000 (Pascal) I didn't get as much use out of it before now. I also got a Maxwell based Nvidia GPU but it too fell under this issue. *sigh*

Fortunately using the hugely bloated kernel amdgpu drm instead of the radeon drm is more stable on my gcn1 video cards. At least the in kernel size of amdgpu is like 20MB and radeon is like 3MB???

video card software compatibility bites the big one...

[rab0171610]

Nvidia GPUs are typically geared toward mid-level to high end gaming. For the most part, it is not a general desktop GPU although there have been some basic GPU's that would be considered that. The NVIDIA Quadro P1000 fell under the category of professional graphics but there were some people that got one on the second hand market and attempted to use it for gaming.

A graphics card that is used intensely or even moderately is not going to last forever. Heat, dust, stray currents, etc. all play their part to slowly degrade chips. My Nvidia graphics cards usually just quit one day and I have to replace them.
This is generally every 3 - 5 years on average. I still have some older 1GB models in a cabinet somewhere collecting dust. They are like museum pieces. I continue to buy Nvidia cards even though I am only an occasional gamer. I don't need it but to me I enjoy the best I can afford, including RAM, Motherboards, and CPU's.

They release newer and faster graphic cards every few years as does AMD. They innovate and try to stay competitive in a global market with a lot of players trying to grab a share of the pie. There is a lot of competition.

Nvidia GPU's are popular in the used and refurbished GPU market for those that want to get their hands on something they could not afford when it was initially released.

The Quadro P1000 was not a great seller and was geared toward corporate workstations. Many of those are in computers that are supported by IT professionals who probably would replace a burned out GPU with old stock on hand or they have already upgraded to something newer as a replacement.

Nvidia usually supports a graphics card for many years, well beyond its normal life span. The Maxwell architecture was release in 2014, over a decade ago. Your graphics cards have all reached the end of their life. If you just need something for general desktop computing, Nvidia is overkill. You will need to invest in something more recent if you choose Nvidia. There is no point in wasting money on an old graphics card if it will be obsolete in a decade.

[eccerr0r]

If on the odd chance it does survive, I'd expect software to continue to work with it.

Granted there was a close call with my old ATI GCN1 cards which is about as old or even older... yet it still works on the amdgpu driver. I did have one GCN1 card fail on me, the R7-250 went belly up.

I still have to say F**K Nvidia and I'm glad I didn't pay (much...less than $5) for it... though pretty much all the video cards I've acquired recently I've paid almost nothing for, after my Radeon 5770 died which I did pay full price for. I am NOT happy with any GPU longevity, both software and hardware... at least ATI software has been longer lived... (Intel chipset/cpu graphics have been an exception, I've had only 1 fail on me and it was bad capacitor plague failure...)

[rab0171610]

If on the odd chance it does survive, I'd expect software to continue to work with it.

I respect your opinion. I can also understand from a financial perspective why Nvidia has to move on and stop supporting old architectures and archaic hardware after nearly a decade. A handful of users who are trying to milk an old graphics card for every last cent it is no longer worth is not where they choose to put their resources. At some point they have to consider the product obsolete and remove their focus. I can respect that as well.

In all fairness to Nvidia, a person who bought an old, used graphics card for 5 dollars is not their customer. The person who originally bought the card 10 years ago is the one who compensated Nvidia financially for the product. The original seller chose not to dispose of the original card when they upgraded. Instead they chose to sell it to another person. I don't see how that person, the second hand buyer, has any grounds for complaint 10 years later. Their dissatisfaction is irrelevant and their grievance has no merit.

[eccerr0r]

OSS it like Intel and ATI and let the true hardware worth/survivorship tell when to stop supporting. F**K nvidia. Never buy new nvidia hardware.

[rab0171610]

Yes, unfortunately users are not entitled to an update to the proprietary drivers indefinitely for their card. They are free to continue using the older driver versions that continue to work with their old hardware.