Install of RTX 5070 Ti problematic on Linux

Graphics / Linux Linux
1

Hello, I just received an RTX 5070 Ti.

My motherboard’s bios is up to date.

I tested Debian 12.10, Devuan 5.0.1, Mint 22.1, Ubuntu 25.04, Fedora 42, Manjaro 24.0.5 and Windows 11.

Only Manjaro 24.0.5 and Windows 11 installed without any problems.

Other operating system installers refuse to start, whether in graphical mode, text mode, or safe mode.

My base installation of Debian 12, installed with an RTX 3080 and updated with the proprietary NVIDIA-Linux-x86_64-570.133.07.run driver before upgrading my graphics card, failed to boot with the RTX 5070 Ti, even though everything worked perfectly with my RTX 3080 ?! Debian freezes during the “loading initial ramdisk” sequence.

I’ve never seen an Nvidia card so difficult to install in Linux ?!

I would like to stay on a Debian-based OS… Can anyone help me ?

1 Like
2

I am not sure if this is the problem, but please note that from Blackwell (50xx) only the open source version of the nvidia driver (570 or 575) can be used. And the GSP firmware must be also activated. So if your system had a previous nvidia card and was using the proprietary version, you may want to change the way the driver is installed. Also I would use the distribution packages, not the .run if possible.

1 Like
3

Not sure if it will help with your 5xxx Nvidia GPU as these are known to be extremely unstable and problematic, but Debian-12 (and so Devuan-5 I guess) is a very outdated system: is uses an archaic 6.1 kernel (mind you, this was released in 2022). I strongly recommend using Debian-13: I’ve been using it for over a year now on my primary laptop with 3090 eGPU and it’s been very stable (I’m using the open drivers from the DC apt repo for Debian 12). It will be officially promoted to stable probably sometime late May or June (see DebianTrixie - Debian Wiki ).

4

@ zebcom

Hello, and thank you for your suggestion.

Unfortunately, the Debian 12.x installation doesn’t work with this graphics card. I really need help resolving this issue (if possible), because even if I could modify my current installation to make it work, if it crashes, I’ll no longer be able to use my machine or resoft it, which isn’t an option for me. This is a production machine…

I could have settled for Linux Mint 22.1 LTS or Ubuntu 25.04 (non-LTS) while waiting for Debian to release a compatible installation image, but these two operating systems, which have just been released, also don’t allow booting from installation ISOs.

This situation is completely unusual for me with an Nvidia card, especially since the rest of my hardware hasn’t changed. In my experience, I should be able to boot an OS with the free driver included in the distributions, then install Nvidia’s proprietary .run driver, but it doesn’t work ;-(

And since the Debian folks don’t seem in a hurry to update their Nvidia drivers, the situation doesn’t look good to me ;-(

I’m afraid I’ll have to return this card to its vendor. Debian is ready for AMD 9070 series GPUs, but not Nvidia 5070 series ?!?

This is what happens when you refuse to collaborate with Linux kernel developers: customers get embarrassed, as do resellers, and ultimately Nvidia themselves…

Best Regards

5

@ morgwai666

Hello, and thank you for your suggestion.

The fact that Linux Mint 22.1 LTS or Ubuntu 25.04 (non-LTS) also don’t work, even though they use very recent kernels that have just been released, is not a good sign…

There’s a major risk that Debian 13 won’t work either, and I can’t just sit around with this card for months, hoping it might be supported by Debian, but without any guarantee that this will be the case.

Additionally, Nvidia’s .run driver works with kernel 6.1, and RTX 5070 GPUs should be fine with it… If they start requiring other updated components, their installation will become much more complicated than that of AMD GPUs ;-( The reason Nvidia is chosen for Linux machines is that their .run drivers have worked in all cases so far…

I’m afraid the only solution for me is to switch manufacturers, unless someone has a brilliant idea to get me out of this.

By the way, I want to point out that in my experience there is no guaranteed working image of Debian Sid (the future Debian Trixie). The only existing ISO images are generated daily in an automated manner by machines, and in my experience they are so buggy that their use/installation is unthinkable.

If I could install Debian 12 and then update the directories to migrate to Debian 13, I would be delighted, but unfortunately it does not work; installing Debian 12 is impossible with this GPU ;-(

Best Regards

6

Note that when I say open source vs proprietary I am not talking about nouveau vs nvidia, but the two architectures of the nvidia vendor driver, one being open-source (the proprietary part runs as a firmware o the GPU). As far as I know if you use the .run from nvidia then you need to specify that you want the open source one.

8

@ zebcom

Don’t worry, I understood correctly :)

I’m afraid that the development of the RTX 50xx series has not followed the usual quality standards regarding their compatibility with Linux, which is problematic… ;-(

1 Like
9

@zebcom

Post Scriptum

I don’t know why Nvidia calls its new proprietary .run driver an open source driver.

To my knowledge, only the kernel headers of this proprietary driver are open source; the other elements of this driver are not, and this can indeed be confusing for some people…

Only Nouveau or Nova driver are really open source.

It reminds me of East Germany when I was young, who bore the name of German Democratic Republic (GDR) even while it was a militaristic dictatorship form USSR.

Have a nice weekend !

10

Three days. That’s how long I spent installing my new Asus Prime RTX 5080 OC card on Linux Mint 22.1. At first, everything froze after 3 minutes. Then I found a problem in the motherboard for the 10th generation Core processor… In the end, all that was left of the entire computer was a case and SSD drives. ) As a result, only the 570-open driver from the ppa repository worked, and that was clearly with errors. I need this entire system for Davinci Resolve Studio. Yes, the driver is clearly not ready yet, sometimes (but rarely) the system can freeze. I think the guys from Nvidia will fix everything in 1-2 months. So when I start rendering at 100-200 fps, I really don’t care about rare freezes. I’ll wait a month or two until the stable release.

11

reading this I feel like I’m in a different world than you guys, though I did get a 5090.

in my experience it was extremely seamless. I installed nvidia drivers from rpmfusion (fedora 41), had them changed to open kernel module a bit ago (had a 3080 before this), then I just slotted my 5090 after a month or two of hunting and it just worked.

It’s basically never “correct” to install the driver directly from nvidia, you should follow your distro’s guide on how to get it working. Additionally, not all distros / driver providers will set you on the open kernel module by default, so you need to make sure that you’re on it.

In my experience, using the proper way to get it working it was less pain than any of the 9070 owners I’ve seen complaining about their cards barely working until a new kernel comes out (and then the newer one, and newer one…)

12

Perhaps because it’s based on the work here?

13

Nvidia now uses an open-source kernel module that sits between the kernel and the proprietary firmware blob that as far as I know exclusively runs on the GPU and GSP. What nvidia has done with the “open-source” module is to move all proprietary/trade secrets aspects to the firmware, fixing an old issue where nvidia was running non-free software on the kernel/CPU, which was object of recurrent debates. Note that AMD does the same: although they rely more on external open-source libraries (Mesa, the open-source vulkan implementation, etc), they use proprietary blobs which contain some features that are not implemented in the free modules. I hope this clarifies.

1 Like
14

Have same issue. Got Ryzen 7 7800X3D and 5050TI, tried every distro, manjaro ubuntu etc with latest kernel 6.14. And properiaty drivers are broken. Open Source works but they are bugged too, got graphic card connected via HDMI to TV and from TV hdmi to Soundbar, sound is not working too. Found only one distro that properiaty drivers works there with that graphic card its Nubara OS 42 (based on Fedora). Hope they will fix the issue since im mostly Arch user

Update:
Forgot to mention that on Manjaro 25.01 with kernel 6.12, firstly ive updated kernel to 6.14 and removed 6.12 (becouse otherwise latest 570.144 drivers was compiling to 6.12 kernel, idk why it was chooseing the lower since higher was installed…) and after installing latest drivers RTX was not detecting HDMI plug in and was not sending video to TV… strange.
But when ive plugged HDMI to motherboard to get video from Ryzen integrated gpu ive checked the driver they was not giving any errors and redering graphic with ‘prime-run vkcube’ was working, so thats strange twice. No Video from GPU Hdmi port but card was working…

15

with a rtx 5070 Ti and rtx 3050 for graphics I have the following errors…

Software I use, such as Gromacs, NAMD, GAMESS…etc wil compile fine with 575 or 570 drivers, and cuda-12.9. However, if I do not use the nvidia-57x-open, the software will not run, always GPU detection erors. When I use these drivers, graphics do not work through the 3050. I am then reduced to a command line terminal, the errors are all through xorg windows problems. If I revert back to nvidia-57x, the graphics work fine, but execution of any software outside of graphics gives device not found errors. Nvidia (sorry I miss posted on the 575 discussion feed) detects the rtx 3050 on either drive set, but only detects the rtx 5070 Ti on the open set. the normal lspci on the open detects both, but again only says nvidia device “0000:01:00.0 VGA compatible controller: NVIDIA Corporation Device 2c05 (rev a1) (prog-if 00 [VGA controller])” With the open, the xorg error is "

[ 578.225] (II) NVIDIA(G0): [DRI2] VDPAU driver: nvidia
[ 578.225] (EE)
[ 578.225] (EE) Backtrace:
[ 578.225] (EE) 0: /usr/lib/xorg/Xorg (OsLookupColor+0x14c) [0x64c2bc60803c]
[ 578.225] (EE) 1: /lib/x86_64-linux-gnu/libc.so.6 (__sigaction+0x50) [0x7cda06045330]
[ 578.225] (EE) 2: /usr/lib/xorg/Xorg (InitOutput+0x4aa) [0x64c2bc4ce07a]
[ 578.225] (EE) 3: /usr/lib/xorg/Xorg (InitFonts+0x1d0) [0x64c2bc48ccf0]
[ 578.226] (EE) 4: /lib/x86_64-linux-gnu/libc.so.6 (__libc_init_first+0x8a) [0x7cda0602a1ca]
[ 578.226] (EE) 5: /lib/x86_64-linux-gnu/libc.so.6 (__libc_start_main+0x8b) [0x7cda0602a28b]
[ 578.226] (EE) 6: /usr/lib/xorg/Xorg (_start+0x25) [0x64c2bc475395]
[ 578.226] (EE)
[ 578.226] (EE) Segmentation fault at address 0x10
[ 578.226] (EE)
Fatal server error:
[ 578.226] (EE) Caught signal 11 (Segmentation fault). Server aborting
[ 578.226] (EE)
"

with the non open the errors from software are all the same an example,
"
CUDA compiler: /usr/local/cuda-12.9/bin/nvcc nvcc: NVIDIA (R) Cuda compiler driver;Copyright (c) 2005-2025 NVIDIA Corporation;Built on Wed_Apr__9_19:24:57_PDT_2025;Cuda compilation tools, release 12.9, V12.9.41;Build cuda_12.9.r12.9/compiler.35813241_0
CUDA compiler flags: -O3 -DNDEBUG
CUDA driver: 12.90
CUDA runtime: 12.90

Running on 1 node with total 24 cores, 32 processing units (GPU detection failed: invalid device ordinal)

"

Interestingly it should have detected at least the 3050, but does not.

I wounder if anyone has a workaround, ie how to get the xorg working with the open driver set, or most likely not, get software runing under the nvidia-575 or 570. I have also tried the cuda modules from nvidia, however they do not support blackwell, and they refer you to the nvidia-570 or 575 open. It is also more problematic, as the new mother boards from almost all vendors do not allow choosing the pci slot anymore, and default to the first slot, as I guess it was necessary to remove 10 lines of assembly code in everyone’s bios. And, graphics wise, new blackwell and nvidia drivers will not go back beyond 2050 cards, making it necessary to have at least a 2050 with any blackwell. Combined, this makes it almost impossible to use commercial gaming GPUs in laboratory settings outside of lone serve set ups, also making it necessary to buy 2x the PCs so people can use graphics as well. It seems weird to me. Literally, really simple fixes are all gone. I can also not find anything on the xorg error, other than the C and cuda compiler are most likely competing for the same memory, with no solutions.

16

@rs277 & @zebcom

The kernel headers in Nvidia’s new drivers are free, but the driver itself is still proprietary. Therefore, Nvidia’s driver is still proprietary, not free, as they apparently want to make it seem by calling it that.

17

I came across these articles in the last few days :

NVIDIA Upstreams Newer GSP Firmware For Open-Source Nouveau Driver

Open-Source NVIDIA Blackwell + Hopper Support Slated For Linux 6.16

In short, they clearly neglected to support their new GPUs under Linux when they were released ;-( This is amateurism made in Nvidia.

I returned the RTX 5070 Ti to its seller and reinstalled my RTX 3080 in my machine. I couldn’t keep a graphics card that I couldn’t reinstall my system with from a standard ISO… In 20 years of Linux I have never had such difficulties when changing GPUs.

18

You are confusing things. Nouveau, and the NVK projects are separate from the open-source nvidia drivers (MIT/GPLv2 license), which is the kernel driver module provided by nvidia for 5070, 5070Ti, 5080 and 5090 from day 1 of the card release.

Nouveau and NVK are still in development and as far as I know not useable for many games, but recently have been also supported by nvidia (providing documentation). Also , the GSP firmware is the binary blob that runs on the card, it is included in the nvidia driver. Note that AMD does exactly the same with a firmware that contains their “secret recipe” which runs on their card.

If you are having issues to install it from the .run file (Blackwell requires the MIT licensed module), I would suggest you use a distribution that provides it as package (e.g. Archlinux). Even if you are not an Archlinux user, their documentation is helpful.

19

@zebcom

No, I don’t think I’m confusing things, contrary to what you say… Nvidia, with its new RTX 5000 architecture, seems to have broken compatibility with the current GNU software “Nouveau” drivers, which are nevertheless essential for installing a base system.

As a result, booting from a standard ISO is no longer possible with text or graphics install, the native VGA mode (640x480) is not supported during base installation with these cards.

The only thing I regret is not having tested the Netinstall version of Debian 13 before returning my RTX 5070 Ti. Maybe it would have worked, I’ll never know. I need a working machine, and I had a deadline to return this card if I wasn’t satisfied. If the Nvidia team sent me one, I’d definitely do some more testing, but I can’t just sit around with a card I bought for $700 that’s unusable for me. In all honesty, I’m extremely disappointed.

I certainly would have had far fewer problems with an RX 9070 XT, which only requires Kernel 6.13, Mesa 25.x, and recent firmware to run. For a while now, I’ve only been buying AMD GPUs for laptops and small home PCs. I’ll definitely do the same for my workstation in the future. This experience has made me lose faith in Nvidia, which sells virtually impossible-to-find products at prices that are far too high! It caused me unnecessary stress, which I could have gladly avoided. It also cost me a lot of time and €50 in return shipping.

To answer your question, I have no difficulty installing Nvidia graphics drivers under Linux, whether GNU or proprietary. Thank you nonetheless for your comments. As I said above, Manjaro was the only Linux distribution I’ve tested out of a total of six that allowed me to boot from it and successfully complete an installation. While Arch and its derivatives are “in vogue” among gamers thanks to their use in Steam OS V2, I personally don’t like these distributions at all. Since they’re rolling releases, they’re ideal if you like dealing with system instability issues, etc, a bit like Windows 11, which isn’t the case for me…

And I hate having to constantly update my system before I can use it. From this point of view, I have not found anything better than Debian… it is a modern, ultra-modular and non-commercial distribution, which will 100% probably still exist in 30 years.

Sorry for my bad English…

Best regards.

20

Finnaly after week of reinstalling fresh Manjaro and Kubuntu 25.04 ive found working way.
First of all dont use X11 session as drivers are broken with it, they works only on Wayland, and ofc on dekstop, becouse on SDDM they dont work too.
So a walkarround way is to install fresh system but dont check autologin, its important. Then on first boot SDDM login screen choose Wayland not X11(it will be saved after first login what you choose), update etc, even install driver but dont reboot. After installation of drivers make sure you setup autologin to account in Wayland session. If u dont do it u will end with no video signal from hdmi (Black Screen) on SDDM login screen so its important to setup autologin to account in Wayland Session before u reboot. Thats all

Tested on Kubuntu 25.04 nvidia-driver-570.153 open, Manjaro 6.14kernel with KDE Plasma nvidia-driver-570.144 open
OS fresh installation ofc

Sry but i cant make logs on X11 since there is no video output signal so whatever i will be doing would be on blind with connected display to RTX 5070Ti hdmi

21

Mine hardware: Ryzen 7 7800X3D RTX 5070TI, motherboard GIGABYTE B650 EAGLE

So working method to get Video Signal on HDMI driectly from RTX 5070Ti and with connected Display to it for me was

Precedure (working on both 570.153 and 570.144 drivers) Tested today (29.05.2025)

  1. Install latest Manjaro KDE (not Gnome, Gnome crashes during driver installation) iso with Open Source drivers. Or Install Kubuntu 25.04 same with Open Source driver (Kubuntu 25.04 have already 6.14 kernel)
  2. Make sure u have at least 6.14 kernel (on Manjaro u must update from 6.12 to 6.14 and then remove 6.12!! otherwise driver installation can compile and build to the lower kernel version which dont work too!!!
  3. In Plasma desktop go to System Settings and find SDDM. Check Autologin user account and choose there Wayland. NOT X11!!. On X11 there is also no video hdmi output signal everywhere
  4. Now on Manjaro terminal write “sudo mhwd -a pci nonfree 0300” or on Kubuntu 25.04 add nvidia driver ppa repository and “sudo apt install nvidia-driver-570-open”
  5. Reboot and enjoy video signal from RTX 5070 Ti HDMI automaticaly logged on your user account Desktop

And yeah before someone asks, Ive added everything, modest fbdev in grub loader, changes nothing.
Adding modules load in initframs changes nothing too. So its not that the drivers loaded too slow becouse ive checked the journal