Tuesday, December 23rd 2025

Arch Linux Drops Support for NVIDIA Pascal and Older GPUs

Linux can be a solid option for anyone looking for an operating system that supports older hardware, but bleeding-edge distributions like Arch Linux often drop support for older hardware sooner than other distributions. This is exactly what happened recently, as the development team behind Arch Linux decided to upgrade the default NVIDIA GPU driver to version 590. This change means that Arch Linux will no longer support NVIDIA GPUs from the Pascal generation and older—this means GeForce GTX 1000 GPUs and older. Somewhat comically, this happened shortly after AMD's Linux kernel drivers extended support to AMD GPUs dating back to 2012.

Arch Linux has also migrated the default NVIDIA GPU drivers to the nvidia-open, nvidia-dkms, nvidia-open-dkms packages. Previously, the GTX 900 series GPUs were the oldest NVIDIA graphics cards supported by the Linux distro, although there are ways to work around the new support limitations. Users with older NVIDIA GPUs can manually uninstall the official nvidia, nvidia-lts, and nvidia-dkms packages and replace them with the nvidia-580xx-dkms drivers from the community-driven AUR (Arch User Repository).
Source: Arch Linux news
Add your own comment

71 Comments on Arch Linux Drops Support for NVIDIA Pascal and Older GPUs

#51
ncrs
LogicIt seems the problem is NVIDIA not wanting to open source the code required for kernel integration.
I'm sure a search on WHY?? will turn up opinions/answers. (Money? Something to hide?)
But I just couldn't be bothered thx to AMD and Intel.

Linux 6.19's Significant ~30% Performance Boost For Old AMD Radeon GPUs

I wonder if NVIDIA realizes that this behavior is influencing the buying decisions/bias of many..?
I, for one, do not recommend, use or buy NVIDIA unless I absolutely have to, out of principle.
Usually there are (GPU encode/decode etc enable) settings, hidden away deep in the relevant software, that make AMD cards price/performance competitive.
(I wonder what is costs NVIDIA? to get these settings hidden and off by default..? :) )
NVIDIA does provide an open source kernel module, but it's only compatible with architectures supporting GSP (the embedded super-firmware) which means Turing (RTX 20xx/GTX 16xx) and newer. However this driver is not yet integrated into Linux upstream and there are significant hurdles preventing it from being upstreamed. Will NVIDIA put in the extra effort? Only they know.

The current open source kernel effort, nouveau, supports Pascal but it can't work properly at full capacity without NVIDIA's firmware files. In the past NVIDIA was aggressively against those efforts, going as far as suggesting to the developers that they should not provide a tool to extract the firmware files from Windows driver packages.

As far as I'm aware the current plan is to deprecate nouveau and combine its efforts with the nvidia-open driver for upstreaming.
Posted on Reply
#52
lexluthermiester
Certain people seem to be struggling with relational context. As such further debate is pointless. I'm out.
Posted on Reply
#53
igormp
ncrsAs far as I'm aware the current plan is to deprecate nouveau and combine its efforts with the nvidia-open driver for upstreaming.
Not really, there's a different driver taking shape to replace nouveau, but it's a parallel effort to nvidia-open:
rust-for-linux.com/nova-gpu-driver

It also only supports GSP-enabled GPUs. Fwiw, nouveau also supports those GPUs with GSP and is even able to run the newest NVK with lots of good support and performance has been improving quite a lot since the project has started.
Posted on Reply
#54
bug
lexluthermiesterSo you're saying(showing) the older drivers will still work. Based on the wording of the announcement, it seemed likely that Arch was pulling support for the older drivers and they would no longer function(properly/at all). Would not be the first time something like this has happened at Arch or in the Linux community.
Just to be clear: Arch did pull the plug, but the 580 line is already migrated to AUR and works fine from its new home. That is what I was trying to show (but didn't make very clear).
Posted on Reply
#55
_roman_
LogicI wonder if NVIDIA realizes that this behavior is influencing the buying decisions/bias of many..?
So many people ignore that and buy on purpose Nvidia. That is not really an argument. I stopped actively reading those forums. Most users bought nvidia even knowing the drawbacks. Just compiling the kernel modules after every kernel upgrade annoys me the most with a nvidia card, was the case also the last time i had an nvidia card in 2023. That is a long term issue.

That Project NOVA has not progressed. I would not advocate or advertise that there are any drivers. There was nothing done two months ago. Nothing. They just decided on the kernel documentation and how to setup the tools and structures. I doubt there was any source code for it. I know about because they are using Rust or something else exotic for it. The huge debatte because they moved away from the classical C.

nouveau in 2023 had no support. and before had no support and i doubt that has changed. A big fancy project which never supported any of my gaming laptops or desktop cards. Similar as coreboot, libreboot whatever. Nouveau is a fancy project with hardly any progress, asks for massive donation of graphic cards but never got anything completed or done. Even when the hardware was outdated and old. For some reason I used very old graphic cards which were most of the time from nvidia several years on the market. 9800m gts / 660m gtx and 960 desktop cards are nice samples in my case to see there was nothing implemented at teh point of time when i looked several times in the owning period how the software support was.
Posted on Reply
#56
bug
LogicIt seems the problem is NVIDIA not wanting to open source the code required for kernel integration.
I'm sure a search on WHY?? will turn up opinions/answers. (Money? Something to hide?)
But I just couldn't be bothered thx to AMD and Intel.
What exactly isn't Nvidia willing to open? This very article is about moving to this: github.com/NVIDIA/open-gpu-kernel-modules
LogicLinux 6.19's Significant ~30% Performance Boost For Old AMD Radeon GPUs

I wonder if NVIDIA realizes that this behavior is influencing the buying decisions/bias of many..?
Of course they do. According to this: pcviewed.com/nvidia-vs-amd-discrete-gpu-market-share/ they're doing it right.
Posted on Reply
#57
igormp
_roman_That Project NOVA has not progressed. I would not advocate or advertise that there are any drivers. There was nothing done two months ago. Nothing. They just decided on the kernel documentation and how to setup the tools and structures. I doubt there was any source code for it. I know about because they are using Rust or something else exotic for it. The huge debatte because they moved away from the classical C.
Upstream development for it surely is slow, but most of the development has been going downstream, which had updates 4 days ago:
gitlab.freedesktop.org/drm/rust/kernel/-/tree/drm-rust-next

There's no reason to try to move everything into upstream for an unstable driver. But yes, it's not meant to be used as a daily driver at ALL.
_roman_nouveau in 2023 had no support. and before had no support and i doubt that has changed. A big fancy project which never supported any of my gaming laptops or desktop cards. Similar as coreboot, libreboot whatever. Nouveau is a fancy project with hardly any progress, asks for massive donation of graphic cards but never got anything completed or done. Even when the hardware was outdated and old. For some reason I used very old graphic cards which were most of the time from nvidia several years on the market. 9800m gts / 660m gtx and 960 desktop cards are nice samples in my case to see there was nothing implemented at teh point of time when i looked several times in the owning period how the software support was.
Nouveau got some really nice upgrades after the GSP became a thing. Its performance is ofc way behind Nvidia's own driver, but it actually manages to run almost everything with NVK:
www.phoronix.com/review/nvidia-nvk-linux-618-mesa-26/10

However, Maxwell and Pascal get the short end of the stick given that they lack the GSP and also face the lack of reclocking due to the missing firmware signing required by nvidia:
www.phoronix.com/review/nvidia-980-5080-linux
Posted on Reply
#58
lexluthermiester
bugJust to be clear: Arch did pull the plug, but the 580 line is already migrated to AUR and works fine from its new home. That is what I was trying to show (but didn't make very clear).
Yeah. Took some digging but I found that info. Still not a great move for Arch, but at least users can still get at the proper drivers for their older cards.
Posted on Reply
#59
bug
lexluthermiesterYeah. Took some digging but I found that info. Still not a great move for Arch, but at least users can still get at the proper drivers for their older cards.
Tbh, I am using Arch ALHP, I had my drivers optimized for x86-64-v3. Because of this move, I lost that optimization. So yeah, not a great move at all.
Posted on Reply
#60
igormp
bugTbh, I am using Arch ALHP, I had my drivers optimized for x86-64-v3. Because of this move, I lost that optimization. So yeah, not a great move at all.
Eh, I don't think this is relevant at all for this specific case. If you were using the regular nvidia package (which I doubt since I assume you were using ALHP's kernel as well), then there's nothing to optimize since it's just a binary blob.
For the nvidia-dkms package, if you really want to have the v3 optimizations, you can just edit your `makepkg.conf` to add the proper cflags that ALPH uses:
somegit.dev/ALHP/ALHP.GO/src/branch/main/flags.yaml

My current config snippet for that looks like so:
CPPFLAGS="-D_FORTIFY_SOURCE=2"
CFLAGS="-march=native -mtune=native -O3 -pipe -fno-plt"
CXXFLAGS="-march=native -mtune=native -O3 -pipe -fno-plt"
# GCC 11 still doesn't support mold
#LDFLAGS="-Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now --ld-path=/usr/bin/mold -fuse-ld=mold"
LDFLAGS="-Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now"
#-- Make Flags: change this for DistCC/SMP systems
MAKEFLAGS="-j$(nproc)"
#-- Debugging flags
DEBUG_CFLAGS="-g -fvar-tracking-assignments"
DEBUG_CXXFLAGS="-g -fvar-tracking-assignments"

And that'd be it, you'd have the same optimizations as before. But again, there's not much to optimize to begin with since the majority of the files in those drivers are pre-built libraries and what's being built is just the kernel shim to talk to those libraries.
Posted on Reply
#61
bug
igormpEh, I don't think this is relevant at all for this specific case. If you were using the regular nvidia package (which I doubt since I assume you were using ALHP's kernel as well), then there's nothing to optimize since it's just a binary blob.
For the nvidia-dkms package, if you really want to have the v3 optimizations, you can just edit your `makepkg.conf` to add the proper cflags that ALPH uses:
somegit.dev/ALHP/ALHP.GO/src/branch/main/flags.yaml

My current config snippet for that looks like so:
CPPFLAGS="-D_FORTIFY_SOURCE=2"
CFLAGS="-march=native -mtune=native -O3 -pipe -fno-plt"
CXXFLAGS="-march=native -mtune=native -O3 -pipe -fno-plt"
# GCC 11 still doesn't support mold
#LDFLAGS="-Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now --ld-path=/usr/bin/mold -fuse-ld=mold"
LDFLAGS="-Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now"
#-- Make Flags: change this for DistCC/SMP systems
MAKEFLAGS="-j$(nproc)"
#-- Debugging flags
DEBUG_CFLAGS="-g -fvar-tracking-assignments"
DEBUG_CXXFLAGS="-g -fvar-tracking-assignments"

And that'd be it, you'd have the same optimizations as before. But again, there's not much to optimize to begin with since the majority of the files in those drivers are pre-built libraries and what's being built is just the kernel shim to talk to those libraries.
I am literally in the middle of an update right and while watching the screen, the fact that the driver not being open source means there's not much ALHP could about it anyway just hit me.

My original assertion is probably true for other AURs, but luckily I only use 3 or 4. 2 of them being the ALHP repos and keyring.
Posted on Reply
#62
igormp
bugI am literally in the middle of an update right and while watching the screen, the fact that the driver not being open source means there's not much ALHP could about it anyway just hit me.
Happens to the best of us.
As an aside, I'm still postponing updating my desktop because I'm too lazy to go through the hassle of switching over from Gnome+X to KDE+Wayland haha
▪uptime
23:10:12 up 119 days, 8:32, 2 users, load average: 0,71, 0,69, 0,48
bugMy original assertion is probably true for other AURs
And so does my tip, just include the proper cflags to your makepkg.conf and you're set for any source package in the AUR.
Posted on Reply
#63
bug
igormpHappens to the best of us.
As an aside, I'm still postponing updating my desktop because I'm too lazy to go through the hassle of switching over from Gnome+X to KDE+Wayland haha
▪uptime
23:10:12 up 119 days, 8:32, 2 users, load average: 0,71, 0,69, 0,48

And so does my tip, just include the proper cflags to your makepkg.conf and you're set for any source package in the AUR.
Fwiw, I could never come to terms with Gnome, KDE continues to work just fine on top of X over here. Tbh, I think KDE works fine on top of Wayland, too. Apps on the other hand, not so much.
Posted on Reply
#64
leonavis
bugFwiw, I could never come to terms with Gnome, KDE continues to work just fine on top of X over here. Tbh, I think KDE works fine on top of Wayland, too. Apps on the other hand, not so much.
Haven't had a problem with KDE/Wayland for a long time now
Posted on Reply
#65
igormp
bugFwiw, I could never come to terms with Gnome, KDE continues to work just fine on top of X over here. Tbh, I think KDE works fine on top of Wayland, too. Apps on the other hand, not so much.
I just got used to Gnome and manage to work my way around it, but them dropping Xorg support just broke my workflows completely, and I don't think it's worth it for me to put up with the issues I had in there and re-do all my workarounds in Wayland.
KDE does work with X for the meantime, but since it's going to be dropping that support in 6.8, and I'm already putting in the effort to change DEs, I thought it would be better to go all in and move to Wayland as well to avoid extra work in the future.
Posted on Reply
#66
bug
leonavisHaven't had a problem with KDE/Wayland for a long time now
I haven't tried for at least 6 months now, but when I did, I recall neither Firefox nor Chromium were stable. And yes, I did set the proper Wayland variables for them.
igormpI just got used to Gnome and manage to work my way around it, but them dropping Xorg support just broke my workflows completely, and I don't think it's worth it for me to put up with the issues I had in there and re-do all my workarounds in Wayland.
KDE does work with X for the meantime, but since it's going to be dropping that support in 6.8, and I'm already putting in the effort to change DEs, I thought it would be better to go all in and move to Wayland as well to avoid extra work in the future.
I still don't understand why the end user should even be aware they're running X or Wayland. I keep contrasting this to the way Apple handled moving their stuff from Carbon to Cocoa :(
Posted on Reply
#67
igormp
bugI still don't understand why the end user should even be aware they're running X or Wayland.
I guess the actual end user that's not really keen in that kind of thing just logs in through their login manager and uses whatever is set as the default for their DE.
I don't have a session manager to begin with and prefer to log in from a tty because I'm a weirdo.
Posted on Reply
#68
bug
igormpI guess the actual end user that's not really keen in that kind of thing just logs in through their login manager and uses whatever is set as the default for their DE.
That's the theory, yes. In practice Wayland broke everything it could and then some, before reaching a usable state.
igormpI don't have a session manager to begin with and prefer to log in from a tty because I'm a weirdo.
Console rulez! But I still log in via SDDM.
Posted on Reply
#69
leonavis
bugI haven't tried for at least 6 months now, but when I did, I recall neither Firefox nor Chromium were stable. And yes, I did set the proper Wayland variables for them.
Strange. Haven't had such problems in a long time. Way longer than half a year. Used Wayland on KDE almost as soon as it was available.
Posted on Reply
#70
bug
leonavisStrange. Haven't had such problems in a long time. Way longer than half a year. Used Wayland on KDE almost as soon as it was available.
It's just my experience, I know many have had better luck. The thing is, there's nothing I want in Wayland anyway, so I gave up trying. It's got proper HDR/color management support recently which I would like to take for a spin. But I don't have a monitor that can do HDR properly, so that will have to wait.
Posted on Reply
#71
TechBuyingHavoc
TheinsanegamerNWhen is the last time you tried it? Core 2 era hardware is getting REAL long in the tooth, and especially the first gen stuff with only 1MB of cache per core has become unusable unless you're moving at 56k speeds. The last gen with higher speeds and 6mb of cache is still usable but is noticeably slow and I wouldnt want to do much more then use Libreoffice with one. A modern $300 PC with a celeron will walk all over one of those old E8000 chips.
Core 2 hardware works just fine if you are using appropriate era software. Office 2003 works just great on my dual core Core 2 Duo laptop, it is very zippy. Old browsers, while a massive security risk and I do not recommend this use case, work just fine on Core 2 Duo hardware. You can substitute with a light, low-resource but modern browser and get close to the same results.

The hardware that is unusable and painfully slow/lagging is Pentium 4 era hardware and earlier.
Posted on Reply
Add your own comment
Nov 21st, 2026 05:57 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts