555 release feedback & discussion

Graphics / Linux Linux
123

Monitors signal cuts out after log in to wayland with driver 555.52.04 and with two AW2721D monitors using DP, and a 2021 Samsung QN65Q70AAFXZA using HDMI takes well over 1 min to load the desktop to a black screen and mouse until I click the mouse and the desktop black screen goes away. Sometimes it doesn’t even get to the desktop and the signal just stays cut til I switch to TTY console and reboot to try again. Display settings reset after logging out and logging back in and after rebooting in both wayland and X11.

If I don’t use the 3rd Samsung tv, logging into wayland still causes both monitors to signal cut but they come back after about 10 seconds then load into wayland with reset display settings every time.

Sleep does not work. Monitors signal cuts but system stays running, can see the cpu voltage, current, and temp fluctuating on my motherboard oled display as if it is still running, but unable to wake the displays back up and end up having to hard shutdown. Can’t even enter TTY.

Also causes breeze dark mode to not work sometimes in chrome based browsers until I switch to breeze light and back to breeze dark again. But the desktop task bar and everything still stays dark. Only the browser glitches out to light mode.

When I am able to get into wayland and can change all my monitors to 240hz and 120hz and enable HDR without it cause the desktop to hang. As long as I don’t log out, or reboot, everything works great everything manages to line up. Very inconsistent though.

These issues don’t happen with the 550 driver. Switched back to 550 and all of these issues are immediately gone.

System Configuration:
Kernel: linux 6.9.3-arch1-1
DM: SDDM
KDE Plasma 6.0.5
kwin-explicit-sync
egl-wayland-git
GPU: Asus Strix Gaming OC RTX 3090 TI
CPU: Intel 13900KS
Mobo: Asus Z690 Maximus Extreme

1 Like
124

555.52.04 is pretty solid for me also using a AW2721D and two VX2768-2KP.

OS: EndeavourOS rolling x86_64
Kernel: Linux 6.9.3-zen1-1-zen
Display (VX2768-2KP): 2560x1440 @ 144Hz
Display (VX2768-2KP): 2560x1440 @ 144Hz
Display (Dell AW2721D): 2560x1440 @ 240Hz
DE: KDE Plasma 6.0.90 (6.1 Beta)
WM: KWin (Wayland)
CPU: AMD Ryzen 9 5900X (24) @ 3.70 GHz
GPU: NVIDIA GeForce RTX 3080 [Discrete]

1 Like
125

555.42.04 is working sort of well with KDE Plasma & Wayland, but noticing 3 issues with my RTX 4080:

  1. Significant frame rate drops when running out of VRAM, probably closely related to or same as VRAM Allocation Issues
    only happening in demanding games and 4K resolution, especially Hogwarts Legacy, monitored using MangoHud when VRAM usage is at 100 %. Framerates dropping from ~ 80 - 100 fps down to 25 or even less, making the game unplayable. Restart required.

  2. When using the open source kernel module, I see lots of messages: kernel: NVRM: nvAssertFailedNoLog: Assertion failed: pEventNotificationList->pendingEventNotifyCount == 0 @ event_notification.c:289. Doesn’t seem to cause any issues though. Not seeing this with the proprietary module, otherwise both modules open and proprietary behave more or less the same.

  3. Turning on VRR sometimes causes a black screen / flickering, no issues with VRR turned off.

126

Browsing SMB causes kioslave5 to throw errors in libnvidia-tls.550.78

This bug was reported against 550.78, but it’s still present in the 555.52.04 on Ubuntu 24.04.

127

555.52.04
Ubuntu 24.04

When using two screens of different resolution (2560 x 1600 (laptop) and 1920 x 1080 (external)) with fractional scaling and switching between display modes with super P (mirror, join displays, external only, built-in), mirror mode has the wrong scaling (too big), and this doesn’t go away after switching away from mirror mode on wayland. Is this an issue with the driver or with wayland?

128

Linux 6.9.3-zen1-1-zen
555.52.4.0, RTX 4070 (12GB VRAM)
KDE6.0.5 / Wayland / Garuda (Arch)
Two monitor setup (2560x1440@120 + 1920x1080@100)

Overall the driver is great, especially on KDE using Wayland. It fixes many issues on KDE Wayland, but there’s some stuff that might need to be looked into:

[drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to allocate NVKMS memory for GEM object

While playing Tarkov (and probably with other games as well, didn’t test) whenever the VRAM usage is high (starting at around ~85-90% utilization), resulting in the image to freeze and in some cases kwin to crash.

Detailed log output 
Jun 11 19:28:15 kwin_wayland_wrapper[2054]: Error getting buffer
Jun 11 19:28:15 kernel: [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00002600] Failed to allocate NVKMS memory for GEM object
Jun 11 19:28:15 kernel: [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00002600] Failed to allocate NVKMS memory for GEM object
Jun 11 19:28:15 kwin_wayland_wrapper[2054]: src/nv_gbm.c:123: GBM-DRV error (nv_gbm_bo_create): DRM_IOCTL_NVIDIA_GEM_ALLOC_NVKMS_MEMORY failed (ret=-1)
Jun 11 19:28:15 kwin_wayland_wrapper[2054]: src/nv_gbm.c:123: GBM-DRV error (nv_gbm_bo_create): DRM_IOCTL_NVIDIA_GEM_ALLOC_NVKMS_MEMORY failed (ret=-1)
  • Experiencing a weird scaling issue on KDE icons in the panel - I’ve reported it to KDE devs in this bug report, but they referred to the Beta driver being the issue here…

Apart from that, great experience!

1 Like
129

Seems like the Xwayland VRAM issue isn’t new.
Sorry to ping you @aplattner but this really needs to be addressed, it’s probably a bug in the driver.

There’s some information about it here, Xwayland uses a lot of VRAM after resizing windows (tested firefox and emacs) (#1617) · Issues · xorg / xserver · GitLab

And I also created this:
Xwayland: GPU memory leak (#1687) · Issues · xorg / xserver · GitLab .
Update: For the nvidia devs, there are some new info in the issue above (1687).

130

I have an issue where opening a campaign in Total War: Warhammer 3 under Proton sometimes causes Xwayland to crash. Runs fine under X11, and the logs show Nvidia driver errors before any other errors. This happened to me on both 555.42 and 555.52, and it also happened simultaneously to me and a friend when joining the same game. It has sometimes seemed to work fine, but then it crashes again. This probably isn’t the only way to provoke the crash, but it’s reasonably consistent.

Fedora 40 KDE and Fedora 40 Kinoite. Kernel 6.8.10. Kwin 6.0.5-3 (patched with explicit sync by Fedora) and Xwayland 24.1.0. Multiple different Proton versions tried(8, 9, 9-GE etc). Running with nvidia-drm.modeset=1 nvidia-drm.fbdev=1 nvidia.NVreg_EnableGpuFirmware=0

Here’s the journal log around the time of the Xwayland crash. I note that there’s a kernel-level error in the nvidia_drm module first, and then kwin logs a syncobj protocol violation, and apps start noticing that Xwayland is dead. This order of events is why I’m here and not in the Xwayland or Kwin bug trackers, but I may be wrong here.

My suspicion is that the driver issue causes the buffer, which is provided by the driver because of fbdev=1, to disappear. Since the buffer disappeared, Kwin throws a protocol violation “explicit sync is used, but no buffer is attached”, which is fatal to Xwayland.

I’ll probably try without fbdev=1 to see if I’m right about the above, especially if the driver error persists but Xwayland stops crashing.

Jun 08 23:37:38 fedora steam[151092]: WARNING: discarding _NET_WM_PID 5 as invalid for X11 window - use specialized XCB_X11_TO_PID function!
Jun 08 23:37:38 fedora steam[151092]: WARNING: discarding _NET_WM_PID 5 as invalid for X11 window - use specialized XCB_X11_TO_PID function!
Jun 08 23:37:47 fedora kwin_wayland_wrapper[4939]: src/nv_gbm.c:123: GBM-DRV error (nv_gbm_bo_create): DRM_IOCTL_NVIDIA_GEM_ALLOC_NVKMS_MEMORY failed (ret=-1)
Jun 08 23:37:47 fedora kernel: [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000a00] Failed to allocate NVKMS memory for GEM object
Jun 08 23:37:47 fedora kernel: [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000a00] Failed to allocate NVKMS memory for GEM object
Jun 08 23:37:47 fedora kwin_wayland_wrapper[4939]: src/nv_gbm.c:123: GBM-DRV error (nv_gbm_bo_create): DRM_IOCTL_NVIDIA_GEM_ALLOC_NVKMS_MEMORY failed (ret=-1)
Jun 08 23:37:48 fedora kwin_wayland_wrapper[4939]: Error getting buffer
Jun 08 23:37:48 fedora kwin_wayland_wrapper[4819]: error in client communication (pid 4819)
Jun 08 23:37:48 fedora kwin_wayland_wrapper[4939]: XWAYLAND: wp_linux_drm_syncobj_surface_v1@70: error 3: explicit sync is used, but no buffer is attached
Jun 08 23:37:48 fedora kwin_wayland_wrapper[4939]: (EE) failed to dispatch Wayland events: Protocol error
Jun 08 23:37:48 fedora goverlay[171116]: The X11 connection broke (error 1). Did the X11 server die?
Jun 08 23:37:48 fedora xwaylandvideobridge[5326]: The X11 connection broke (error 1). Did the X11 server die?

Here’s a dmesg log from a separate crash:

[   18.730787] systemd-journald[863]: File /var/log/journal/20b89bf7b2d4455094a72d0dbe3fd0c5/user-1000.journal corrupted or uncleanly shut down, renaming and replacing.
[  229.193536] nvidia 0000:1c:00.0: Using 47-bit DMA addresses
[  244.470257] [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00001c00] Failed to allocate NVKMS memory for GEM object
[  244.470329] [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00001c00] Failed to allocate NVKMS memory for GEM object

The driver errors here seem similar to some of the other posts here, like the two directly above me.

1 Like
131

This issue occurs in various games that want to use a lot of VRAM, presumably because the nvidia driver doesn’t do memory management very well. On my 10GB card I usually have to close as many applications as I can to be safe from crashes.

1 Like
132

Hello gents,
How does PowerMizer work in version 555? Does it still waste electricity when it’s not needed? Or is there an option again to lock the low frequencies, like in 525.125?
Thanks!

133

sudo apt install nvidia-driver-555 caused my Ubuntu 22.04 to boot into a complete black screen even in recovery mode, which caused me to unable to access the virtual console, which means ctrl + alt + F2 doesn’t even work. I had to boot with a LIVE usb and mount my drive and do sudo apt purge nvidia-* to remove 555 to recover. Eventually my machine come back to life again after reinstalling the previously working 550 with sudo apt install nvidia-driver-550. However, no matter I try sudo apt install cuda-12-5 or sudo apt install cuda-12-4 or sudo apt install cuda-12-3, it all wants to remove my 550 driver and install my 555 driver. Is there a way to install cuda with the 550 driver so I can get back to the original working state on my machine?

134

Are you in a laptop right? What is the exact problem you have? You want to get your GPU suspended when is not in use to save battery?

135

Related: Non-existent shared VRAM on NVIDIA Linux drivers

136

Correct, I’m on i7 laptop with legacy GTX 1060, but it’s more than enough for common use.
Starting from 525.125 Nvidia drivers behaviour of PowerMizer changed and GPU power consumption become too aggressive. Even with adaptive configuraton it switches on maximum performance level and heats up the card pretty good and unnecessarily burns power even if start nvidia-settings (which is critical when on battery). With other apps like browser, messanger, development app for working or just some classic game where graffic performance is not necessury - driver keeps max performance constantly.
More about this problem here and here
Hope Nvidia get this fixed in new driver versions. For now, as many other people, we are limited with 525.125 and have to modify new kernels to setup it.

137

I am still getting lower performance after a system suspend to RAM and resume with the 555.52.04 drivers and my Nvidia 3090 (Desktop). It is most prevalent in Hellblade 2, but recently I found out that it also happens in Counter-Strike 2. On the Mirage map, I usually get 140-160 fps on a fresh boot. If I suspend and launch the game again, I only get 90-100fps.

2 Likes
138

Have you ever tried using PCI-Express Runtime D3? It power offs your GPU when is not in use idling it at 0W PRIME - ArchWiki

If you are using Wayland you will need to apply PRIME - ArchWiki (Some programs have a delay when opening under Wayland) because NVIDIA drivers are dumb and stupid. More info here: NVIDIA Discrete GPU/DGPU powers on on every program that opens. / Laptop Issues / Arch Linux Forums

I have a 3070 Max-Q on my laptop but I’m using the propietary drivers so this should work for you too.

139

Hello
I had similar issue with 550 until I switched on Hybrid mode in BIOS

140

Setup:

  • OS: Arch Linux, kernel 6.9.5, KDE 6.1, Wayland
  • GPU: ASUS ProArt 4080 Super OC
  • Monitor: Samsung Odyssey G8 OLED

Problems with the current driver (550.xx):

  • HDMI cause tearing, while DP works ok.
  • If I use DisplayPort HDR does not work: colors are washed out for some reason.

Updating to 555 didn’t fix anything:

  • HDMI now doesn’t work at all. Monitor just loses signal.
  • HDR with DisplayPort also doesn’t work, colors are washed out just like with 550.

I tried both nvidia and nvidia-open packages.

141

144 hz is set as 29 hz for me using KMS. All other refresh rates are set normally. Using: Linux 6.9.5, RTX 4070, version 555.52.04.

Repro steps:

  1. Set video kernel parameter: video=1920x1080@144
  2. Reboot
  3. Observe monitor is set to 29 hz in TTY, not 144 hz like it should
  4. Start a Wayland/X session, monitor is still set to 29 hz (even when mode is explicitly set by compositor)
  5. Change refresh rate to 120 hz, it changes to 120 hz.
142