NVIDIA has recently released the second beta for the 560 Driver.
After a lot of testing, we have decided to push the 560 driver to our repository. Cuda will be with version 12.6 provided, to keep compatibility with these drivers.
Here you can find the changelog:
* Updated nvidia-installer to select the NVIDIA open GPU kernel modules by default on systems with GPUs that support both the proprietary and open kernel modules.
* Fixed a bug that caused GPU driver installation to fail when the system used alternate implementations of the 'tr' utility, such as from the busybox or toybox projects.
* Fixed a bug that could cause the wrong image format to be used for render pass image clears in Vulkan applications when using a VkImage created with VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT. This could lead to rendering corruption, as described in issues such as:
https://github.com/doitsujin/dxvk/issues/3961
* Fixed multiple issues that could cause crashes or unexpected behaviors when re-creating an NvFBC capture session.
* Added support for EGL_KHR_platform_x11 and EGL_EXT_platform_xcb on Xwayland.
* Fixed a bug that could cause some displays to appear multiple times in the nvidia-settings display layout configuration page on systems with multiple GPUs.
* Added a PipeWire backend to NvFBC that allows it to work on the Wayland compositors that support screencasting via XDG Desktop Portal. This new interface will be available through an upcoming Capture SDK release.
* Added support for multiple concurrent clients to NvFBC direct capture.
* Added reporting of Vulkan information to nvidia-settings control panel.
* Compiling nvidia-settings from source now requires Vulkan header files to be available.
* Support DRM-KMS explicit synchronization via the IN_FENCE_FD mode setting property.
* Support VRR (Variable Refresh Rate) for Wayland on pre-Volta GPUs.
* Added support for Variable Refresh Rate on notebooks with the open kernel modules.
* Updated glXWaitVideoSyncSGI() to be more efficient. This reduces frame stutter in some KDE configurations with GSP offload.
* Fixed a bug that caused OpenGL triple buffering to behave like double buffering.
NVIDIA suggest to use the Open Module as default, for supported cards.
Supported Cards are any, which are newer then the 20xx series. These cards have the GSP Processor, which is used by the open driver.
The closed source module will get less Quality Testing for future releases and nvidia wants to focus on the open source module.
Users, which have a supported card can simply switch with following command to the open source modules:
sudo pacman -Syu linux-cachyos-nvidia-open
In case you have more kernels installed, replace these also with the *-open at the end.
Example:
In case you have the linux-cachyos-bore-nvidia package installed, you can switch with running sudo pacman -S linux-cachyos-bore-nvidia-open
New Installation, will start with the upcoming release automatically, if supported. This has been implemented already in our hardware detection and will be rolled out soon.
Nice, thanks for testing this. Looking pretty good for me.
I can run and play Hunt Showdown without any issues.
On 15 August Crytek will upgrade the Engine and rename the game to Hunt: Showdown 1896
I hope the game will still run fine on Linux after the big upgrade.
Switch back to the closed version of this driver. the open modules still have a lot of stutter in certain games, mainly Overwatch 2 which is one of my main games atm.
The open driver seems to ignore my GSP disabled config and with GSP enabled there is still a noticeable amount of stutter. It is indeed much better than before, I’d say around 70-80% better but its still very much present and can be felt a lot in such a snappy game like OW2.
As soon as I went back to the closed version the game immediately felt perfect again. Still some work needs doing for these drivers.
When trying to update, I get a dependency conflict. The system wants to update to NVIDIA 560.31.02, but it conflicts with the current 555.58.02 version.
Running sudo pacman -Syu gives this error:
~ > sudo pacman -Syu
:: Synchronisation des bases de données de paquets…
cachyos-v3 est à jour
cachyos-core-v3 est à jour
cachyos-extra-v3 est à jour
cachyos est à jour
core est à jour
extra est à jour
multilib est à jour
:: Début de la mise à jour complète du système…
résolution des dépendances…
recherche des conflits entre paquets…
erreur : la préparation de la transaction a échoué (la satisfaction des dépendances a échoué)
:: l’installation de nvidia-utils (560.31.02-1) casse la dépendance « nvidia-utils=555.58.02 » requise par nvidia
Trying sudo pacman -Syu nvidia results in:
~ > sudo pacman -Syu nvidia
:: Synchronisation des bases de données de paquets…
cachyos-v3 est à jour
cachyos-core-v3 est à jour
cachyos-extra-v3 est à jour
cachyos est à jour
core est à jour
extra 7,3 MiB 22,0 MiB/s 00:00 [---------------------------------------------] 100%
multilib est à jour
avertissement : nvidia-555.58.02-13 est à jour -- réinstallation
:: Début de la mise à jour complète du système…
résolution des dépendances…
recherche des conflits entre paquets…
erreur : la préparation de la transaction a échoué (la satisfaction des dépendances a échoué)
:: impossible de satisfaire la dépendance « nvidia-utils=560.31.02 » requise par lib32-nvidia-utils
:: impossible de satisfaire la dépendance « nvidia-utils=560.31.02 » requise par linux-cachyos-nvidia
How can I resolve this?
Kernel: Linux 6.10.3-3-cachyos
DE: GNOME 46.4
WM: Mutter (Wayland)
CPU: Intel(R) Core™ i7-10700K (16) @ 5,10 GHz
GPU: NVIDIA GeForce RTX 3080
This driver needs more work then because the stutter with this GSP firmware enabled makes certain games unplayable like I said OW2 is a stuttering mess. So I moved to the closed version of this driver and it works flawlessly.
The stutters are vastly improved though on the new open driver, just needs more optimization to get it performing like closed driver.
‘gdm is blocking nvidia wayland, whyever they do.’
Okay, I get it, @ptr1337. I will switch to KDE Plasma!
The only difference is that Plasma takes a bit more video memory than Gnome, which sometimes makes a difference when it comes to the horrible error message ‘CUDA out of memory’.
I have some screenshots of nvtop in Plasma+Wayland, Gnome+x11, and NobaraGnome+Wayland for comparison.
Hi @ptr1337 . I tried this, and the cog wheel appeared at the gdm on reboot. However, when I try to log in after choosing Wayland, there is a black screen briefly, and I am back at gdm. I tried the steps in Arch Wiki, but they did not work. I found that article so convoluted due to my lack of understanding of Arch Linux at a deeper level. It is okay. The good old x11 is fine by me.
