GTX 1650 crashes on Fedora 44 Wayland - NVRM RPC Timeout & TLB Invalidation Failures

Ask Fedora
,
1

Hi everyone,

I am experiencing hard GPU freezes on Fedora 44 under Wayland while gaming via Proton/DXVK. The system locks up entirely, and nvidia-smi drops the card (No devices were found) until I perform a full cold reboot.

System Specs

  • OS: Fedora 44 (KDE Plasma Desktop Edition)
  • Compositor: KWin Wayland
  • GPU: NVIDIA GeForce GTX 1650 Mobile (GM107/Turing hybrid)
  • Driver: 610.57.04 (Proprietary package loading open kernel modules backend)
  • Persistence Mode: Off

Core Error Logs (dmesg / messages)

kernel: [drm:__nv_drm_semsurf_wait_fence_work_cb [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to register auto-value-update on pre-wait value for sync FD semaphore surface
kernel: NVRM: GPU0 kgmmuInvalidateTlb_GM107: TLB invalidation failed waiting for prior invalidate (status=0x0000000f), vaspaceFlags 0x101, scope 0x2, GFID 0
kernel: NVRM: _issueRpcLarge: rpcSendMessage failed with status 0x0000000f for fn 76!
kernel: NVRM: nvAssertFailedNoLog: Assertion failed: 0 @ rpc.c:2095
nvidia-powerd[805]: ERROR! Failed to get AC Line status

Proposed Workaround By AI

The issue appears to be a race condition/memory state loss between Wayland’s explicit sync protocols and the driver’s power state transitions (Persistence-M: Off).

I have been advised to create a user modprobe override to enforce VRAM preservation to a persistent disk path (btrfs partition with 325GB free):

# File: /etc/modprobe.d/nvidia-power-management.conf
options nvidia NVreg_PreserveVideoMemoryAllocations=1
options nvidia NVreg_TemporaryFilePath=/var/tmp

Followed by a sudo dracut --force image rebuild.

Questions for the Team

  1. Is this specific modprobe workaround trusted and safe to address this exact semsurf_wait_fence / rpcSendMessage timeout loop on Fedora 44?
  2. Are there any known side effects regarding early boot display or performance on this specific architecture?
  3. Is there a cleaner upstream patch or alternative I should implement instead?

Thanks for your time and feedback!

2

Nonsense! pls read the nvidia driver documentation instead of wasting time with AI. Both options change settings how the system saves VRAM while suspended.

3

Yeah, that AI slop looks like nonsense.

First, upgrade to the latest driver and see if this goes away. Latest is currently 615.71.09.

Aside from that, I doubt this issue is something most folks here could help with. You should open a thread on the NVIDIA developer forum: Linux - NVIDIA Developer Forums
Be sure to follow their best practices to report the issue. You’ll need to provide a bug report file, instructions here: »»»»»»»»»» If you have a problem, PLEASE read this first «««««««««« - Linux - NVIDIA Developer Forums
There’s also a feedback thread for the driver where you can reference your issue.

EDIT: I see 610 is the latest driver in the rpmfusion-nonfree-nvidia-driver repo, for some reason. Presumably that is the repo you are using. You may consider configuring the full RPM Fusion nonfree repo instead, as per these instructions: https://rpmfusion.org/Configuration

4

Thank you for your response. And yes, I tried to go though the official documentation. However, I am new to Linux, so far I could not find anything that relates to my issue. I am mostly relying on AI or Reddit. But I thought asking to the forum would give me some correct and trusted answers.

Regarding the issue, I would like to provide more details:

5

Thank you for your response. Based on your advice, I have created a thread for the issue on NVIDIA Developer Forums.

As for the RPM Fusion nonfree repo, I followed the configuration link, and I have installed the full RPM Fusion nonfree repo.

Before installation:
max@undefined:\~/Max/Scripts$ sudo dnf repo list | grep rpmfusion 
**rpmfusion**-nonfree-nvidia-driver                RPM Fusion for Fedora 44 - Nonfree - NVIDIA Driver 
**rpmfusion**-nonfree-steam                        RPM Fusion for Fedora 44 - Nonfree - Steam 

After installation:
max@undefined:\~/Max/Scripts$ sudo dnf repo list | grep rpmfusion 
**rpmfusion**-free                                 RPM Fusion for Fedora 44 - Free 
**rpmfusion**-free-updates                         RPM Fusion for Fedora 44 - Free - Updates 
**rpmfusion**-nonfree                              RPM Fusion for Fedora 44 - Nonfree 
**rpmfusion**-nonfree-nvidia-driver                RPM Fusion for Fedora 44 - Nonfree - NVIDIA Driver 
**rpmfusion**-nonfree-steam                        RPM Fusion for Fedora 44 - Nonfree - Steam 
**rpmfusion**-nonfree-updates                      RPM Fusion for Fedora 44 - Nonfree - Updates

Could you please review if I have installed it correctly? And also, could you please tell me how this could be helpful to resolve my issue?

Edit:
I had some updates in my “Discover” app, and after installing all the updates, my driver also got a new update:

max@undefined:\~$ nvidia-smi 
Sat Sep 19 10:10:33 2026        
+-----------------------------------------------------------------------------------------+ 
| NVIDIA-SMI 615.71.09              KMD Version: 615.71.09     CUDA UMD Version: 13.4     | 
+-----------------------------------------+------------------------+----------------------+ 
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC | 
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. | 
|                                         |                        |               MIG M. | 
|=========================================+========================+======================| 
|   0  NVIDIA GeForce GTX 1650        Off |   00000000:01:00.0 Off |                  N/A | 
| N/A   43C    P8              3W /   30W |       7MiB /   4096MiB |      0%      Default | 
|                                         |                        |                  N/A | 
+-----------------------------------------+------------------------+----------------------+ 
 
+-----------------------------------------------------------------------------------------+ 
| Processes:                                                                              | 
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory | 
|        ID   ID                                                               Usage      | 
|=========================================================================================| 
|    0   N/A  N/A            2020    C+G   /usr/bin/kwin_wayland                     5MiB | 
+-----------------------------------------------------------------------------------------+

I will now try to play the game again and check if it still crashes. Thank you!

6

At times there is a delay between having the packages available in the rpmfusion-nonfree-updates repo and when they are available in the rpmfusion-nonfree-nvidia-driver repo. It seems this is one of those instances and after configuring the rpmfusion-nonfree-updates repo you were able to update to the latest available driver version.

7

what exactly happens when you run into this issue? Does only the game crash and you can continue using the current desktop session? Or is your session also affected by this?

you could try disabling explicit sync or try replicating with X11.

__NV_DISABLE_EXPLICIT_SYNC=1 disables explicit sync for a program . Try adding that to the games command line .

8

Looks good to me. And you got the updated driver - great!
One small thing you may want to do is to disable the rpmfusion-nonfree-nvidia-driver and rpmfusion-nonfree-steam repos since you now have rpmfusion-nonfree which is a superset of them.

9

The steam packages within the rpmfusion-nonfree-steam repo is also contained within the rpmfusion-nonfree or rpmfusion-nonfree-updates repos. If the user chooses the steam repo can be disabled though doing so is unnecessary.

Disabling the rpmfusion-nonfree-nvidia-driver repo is unnecessary since dnf looks at both that repo and rpmfusion-nonfree-updates and automatically selects the latest packages for installation or updates.

In both cases, dnf chooses the latest version of packages to install and those repos do not conflict.

10

Sure. It’s not a big deal either way but it is better to not have a redundant repo query when doing updates.

11

For the purist limiting the repos may matter a lot, but the only extra time & data involved is limited to the metadata download from the repo.

That is usually only time consuming if the --refresh option is used since there is an automatic periodic refresh of metadata that keeps the system info fairly current. Even when using refresh it only takes a few seconds to download the metadata from about 12 repos I have configured.

Thus, it seldom has any impact for the average user. New users should not even worry about that unless it introduces a noticeable delay into the process.

This part of the discussion is off topic for the issue posed by the OP so should not continue.