I can confirm this and have already opened a separate issue about it. I first noticed it with the 590 drivers and also came across a GitHub issue in proton where people are having the same artefacting.
6017526 Crimson Desert still soft locks upon completing shader compilation part of loading (field of cubes) with driver 595.58.03
Someone figured out a workaround to this bug using vkd3d-protonās non-merged VK_EXT_descriptor_heap. See more information here: [Bug] Crimson Desert only works on the 595 drivers for Nvidia Series 5000 cards (Blackwell) using the VK_EXT_descriptor_heap (PR #2805) Ā· Issue #2914 Ā· HansKristian-Work/vkd3d-proton Ā· GitHub
2 Likes
thanks, it resolved my Xid 109 CTX SWITCH TIMEOUT issue on The Last of Us part II as well (Bazzite OS) ā¦ for now
1 Like
Elden Ring Nightreign freezes on world load with RTX 5090 + 595.58.03 driver
System Info:
- OS: Nobara Linux 43 (KDE Plasma)
- Kernel: 6.19.10-201.nobara.fc43.x86_64
- GPU: NVIDIA GeForce RTX 5090
- Driver: 595.58.03
- CPU: AMD Ryzen 9 7950X
- RAM: 93GB
- Desktop: Wayland / KDE
Proton versions tested: Experimental, Hotfix, 9.0, GE-Proton10-34
Issue:
Game loads into Roundtable Hold without issues. When attempting to load into the main game world, the game freezes (GPU drops to 0% utilization while CPU spins at ~260%). Display reverts to showing previous menu/area. Must force-kill to exit.
Workarounds attempted (none worked):
- VKD3D_CONFIG=no_upload_hvv
- taskset -c 1-31 (CPU 0 affinity fix)
- Cleared shader cache
- Disabled VRR
- Windowed/Borderless modes
- CPU power limit to 99%
- VKD3D_CONFIG=single_queue
Logs show: Flood of Invalid alignment 4096 for buffer resource warnings from vkd3d-proton.
Note: This appears related to known RTX 50-series + Nightreign issues also affecting Windows users.
Why dont you try CachyOS-Proton with PROTON_VKD3D_HEAP=1 just to see if it helps? I am actually curious.
1 Like
Try with my workaround for Elden Ring (not Nightreign):
Regards
Still getting the Xid 109 crashes in Resident Evil Requiem when using /WineDetectionEnabled:False and enabling raytracing. Started happening with the 595 beta and is now in the 595.58.03 release.
Confirming the heaps fork is fixing Crimson Desert for me on Blackwell. And when I said soft lock above, I really meant a hard lock. And it might have been a VRR issue instead.
Speaking of which, I have a different VRR issue, going back to 580.xx, possibly as far back as Iāve tested, which may go back farther. Basically, if I enable adaptive sync / VRR on a wlroots compositor, alt-tabbing out of a Wine Wayland app causes its Vulkan session to freeze, locking up the Wayland session for the app. The window can no longer be moved or manipulated in any way. The audio keeps playing, though.
Workaround is either to disable Vsync on the compositor having Tearing enabled, or to disable Adaptive Sync in the compositor. Simply disabling it on the output with wlr-randr does not fix it.
Edit: Further workaround: Setting Adaptive Sync to full-time enabled also fixes it. Apparently the act of toggling Adaptive Sync on or off breaks Vulkan sessions?
Edit 2: And just for completenessā sake, I tested Crimson Desert with the workaround enabled, it still hard locks if PROTON_VKD3D_HEAP=1 is not set, with an Xid 109 in the journal.
3 Likes
I have compiled the list of some generic issue:
1- There is a performance penalty on using native wayland implementation of SDL or wine-wayland using nvidia driver, mention some tests here, Performance Loss wine-walyand / SDL-wayland - #7 by Syntist
2- Tearing Support doesnāt work if the app is running on xwayland, it doesnāt imedidata flip on kde even if you have tearing enabled, but work if the app is run in native wayland and vulkan is being used. I mentioned the detail here on how i tested it, and it visually seeable with slow motion camera, NVIDIA BUG, KDE Wayland Games are Force Vsynced - #51 by Syntist
3- Direct Scanout is broken for me on my laptop, maybe cause of laptop configuration AMD/Nvidia, I mention more detail here Direct Scanout not working on Hybrid Laptop (AMD/Nvidia) , Already contacted KDE/Mesa and also linux kernel, it something has to be dealt on the driver level, according to zamundaaa (Xaver)
nvidia-bug-report.log.gz (591.2 KB)
Testing the 595 branch because of Resident Evil Requiem.
595.58.03 works with path tracing now. Was not working on 595.45.04 so kudos for that.
But when I try to play Ori and the blind forest it crashes. This was not happening on the 580 branch.
I know Ori is an old game, but is the development only focussed on new games?
Hi everyone,
I am experiencing severe stuttering and massive frame drops on my laptop equipped with an NVIDIA GeForce RTX 3050 Ti Laptop GPU. The issue occurs specifically when the VRAM usage reaches its limit. Instead of gracefully handling the memory pressure, the system logs are flooded with NVRM error messages, and the application performance collapses. The problem is not game-specific; it occurs in every title I run as soon as the VRAM usage reaches its physical limit. Currently, the laptop is only capable of running lightweight indie games.
System Environment:
-
GPU: RTX 3050 Ti Laptop (4GB VRAM)
-
Driver Version: 595.58.03
-
OS: EndeavourOS
-
Kernel Version: 6.19.9
When the game/application fills the 4GB of VRAM, the driver starts reporting NV_ERR_INVALID_ARGUMENT. This leads to an endless loop of assertions in gpu_vaspace.c. Iām attaching a log where you can see this spam for yourself. Iād be very grateful if you could look into my issue.
Thanks in advance.
nvidia-bug-report.log.gz (341.1 KB)
1 Like
Hmm. Iām having same issues on a 5090 Laptop GPU with 595.58.03 as I had a while back with 580.95.05, where external monitor (4K, 240Hz, HDMI) sometimes doesnāt come back on when resuming from suspend, and when it does turn back on, it turns on with 60Hz instead of 240Hz. These issues were fixed with 590.x, but now theyāre back with 595.x.
Works well for me with regular proton-cachyos-10 and PROTON_ENABLE_WAYLAND=1 %command%
I also experience the freeze with Steamās default proton experimental.
It seems that 595 drivers kind of broke Elden Ring for me under Steam and Proton on Arch Linux.
When I start the game and try to load a save file the game just hangs mid load for a short time and then crashes. It didnāt happen with the previous 590 drivers. I tried using proton-cachyos with PROTON_VKD3D_HEAP=1 flag and even though it fixed the crush it looked like the performance dropped significantly and some visual glitches appeared.
What fixed this for me is using an older Proton 8.0.5. The game runs as usual.
Other games like Hell Divers 2 and Cyberpunk run under Proton Experimental just fine.
My System:
OS: Arch Linux x86_64
Kernel: Linux 6.19.10-arch1-1
Host: MS-7E62 (2.0)
Motherboard: MAG B850 TOMAHAWK MAX WIFI
CPU: AMD Ryzen 7 9800X3D (16) @ 5.27 GHz
GPU: NVIDIA GeForce RTX 5070 Ti [Discrete]
Display (Q27G4Z): 2560x1440 in 27", 240 Hz [External]
DE: GNOME 49.5 (240HZ, No VRR)
WM: Mutter (Wayland)
Proton Version: Proton Experimental
nvidia-smi:
Ā±----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.58.03 Driver Version: 595.58.03 CUDA Version: 13.2 |
Ā±----------------------------------------Ā±-----------------------Ā±---------------------+
| 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 RTX 5070 Ti Off | 00000000:01:00.0 On | N/A |
| 0% 38C P3 24W / 300W | 932MiB / 16303MiB | 1% Default |
| | | N/A |
Ā±----------------------------------------Ā±-----------------------Ā±---------------------+Ā±----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 2557 G /usr/bin/gnome-shell 185MiB |
| 0 N/A N/A 2899 G /usr/bin/Xwayland 4MiB |
| 0 N/A N/A 5458 G /usr/bin/alacritty 38MiB |
| 0 N/A N/A 5767 G ./steamwebhelper 30MiB |
| 0 N/A N/A 5799 C+G ā¦am/ubuntu12_64/steamwebhelper 8MiB |
| 0 N/A N/A 6753 G /usr/lib/firefox/firefox 385MiB |
| 0 N/A N/A 12356 C+G /usr/bin/nautilus 43MiB |
| 0 N/A N/A 13995 C+G /usr/bin/gnome-text-editor 64MiB |
Ā±----------------------------------------------------------------------------------------+
Tried to gather logs right after the crash. Hope it helps!
Big thanks for all the work!
nvidia-bug-report.log.gz (334.8 KB)
1 Like
Both Horizon Forbidden West and Zero Dawn Remastered have very unstable performance in the low 50s with going up/down the more you move. This was never an issue and the games ran very smooth.
If it helps the proton log indicates the following:
`14827.971:012c:0130:warn:vkd3d-proton:d3d12_command_list_update_graphics_pipeline: No graphics pipeline bound, skipping draw.
14827.971:012c:0130:warn:vkd3d-proton:d3d12_command_list_DrawInstanced: Failed to begin render pass, ignoring draw call.
14827.972:012c:0284:warn:vkd3d-proton:d3d12_low_latency_device_SetLatencyMarker: PRESENT_START_NV is non-monotonic 90920000 < 90940000.
14827.973:012c:0130:warn:vkd3d-proton:d3d12_command_list_update_graphics_pipeline: No graphics pipeline bound, skipping draw.
14827.973:012c:0130:warn:vkd3d-proton:d3d12_command_list_DrawInstanced: Failed to begin render pass, ignoring draw call.`
Operating System: CachyOS Linux
KDE Plasma Version: 6.6.3
KDE Frameworks Version: 6.24.0
Qt Version: 6.11.0
Kernel Version: 6.19.10-1-cachyos (64-bit)
Graphics Platform: Wayland
Processors: 16 Ć AMD Ryzen 7 5800X3D 8-Core Processor
Memory: 32 GiB of RAM (31,2 GiB usable)
Graphics Processor: NVIDIA GeForce RTX 3090
After update to 595.58.03, the NVIDIA Proprietaryās Licenseās version becomes unable to detect 5070Ti on my computer again.
And both MIT/GPL and NVIDIA/Proprietaryās Licenses still has screen brightness control issue in NVIDIA Optimus Mode on the HP OMEN Note Book.
Hope these problem could be solved in the formal versionš
The 50 series cards never worked with CLOSED modules (which Iām guessing youāre referring to, since both CLOSED and OPEN kernel modules are proprietary).
Besides from 590+ itās not possible to even install them with closed, I believe? Regardless, you should reinstall your drivers as open kernel modules using your distroās suggested install method.
2 Likes
They are still published, at least for Debian: https://developer.download.nvidia.com/compute/cuda/repos/debian13/x86_64/cuda-drivers_595.58.03-1_amd64.deb
They are useful in cases if someone canāt / doesnāt want to use GSP firmware (open flavor cannot function without it). Only on archs older than Blackwell of course, as you pointed.
I have the same issue on Fedora with KDE. Iām using this adapter specifically:
It does however work with an older, non-4k DP-to-HDMI cable I had lying around. (And the adapter above used to work with 590.xx driver.)
1 Like
I have been testing Starfield once again, and Iām noticing 90% GPU usage, yet, way lower FPS compared to Windows (e.g Akila City). Feels like a CPU bottleneck except my 9800X3D should be more than enough on both platforms.
I have noticed though that my power draw in Linux is 190W compared to 360W in Windows, despite similar GPU usages. Could this be a driver bug, or a VKD3D problem?