Starfield no longer crashes on an RTX 2080 Ti after the release of the Vulkan beta driver 535.43.10, however performance was incredibly poor and virtually unchanged from version 525. With the release of 535.43.11, the framerate has increased drastically and is smooth and stutter free. All testing was performed on the nv-dgcc-enable branch. However, this behavior appears vary between GPU architectures. Some users with Pascal cards report that they are still able to launch the game on 535.43.11 but the same cannot be said for Ampere.
Additionally, there appears to be artifacting with certain objects under certain conditions that was not present during 525. For example, the large rocks on the horizon as well as the metallic gas container right side rapidly flicker in and out when moving the camera around, making it tricky to capture a screenshot of:
Starfield v1.7.33
Kernel: 6.5.5
GPU: RTX 2080 Ti
Driver: 535.43.11
Wine version: GE-Proton 8.16
VKD3D-Proton version: https://github.com/HansKristian-Work/vkd3d-proton/commit/01f62c8
Some additional logs on Ampere configuration. Double-checked that DXVK (or rather, its DXGI implementation) does not spoof NV GPU as an AMD one.
log_nvapi_enabled_nvidia_hide_disabled_vkd3d_210_dxvk_23.log
I also get instant GPU hang on 535.43.11 on 3070. I've reported this to NV.
The same on 4080.
The same on 4080.
In my case game hangs just after pressing the load key.
Anybody on 40 series ?
I got instant GPU hang with following Xid crash once I pressed continue in 535.43.11 with a 3080.
NVRM: Xid (PCI:0000:09:00): 109, pid=207127, name=Starfield.exe, Ch 000000ee, errorString CTX SWITCH TIMEOUT, Info 0x2c073
In my case, with 535.43.13 game is no longer crashing, but performance improvements are:
So far, with the config above, the only real difference is frame-time pacing -- it is stuttering noticeably less:
Whilst difference is not that big in terms of sheer numbers, the stuttering is less noticeable and persistent, even on an action-heavy scene like this.
With that said, GPU is still not being utilized properly -- as in, when I decrease internal resolution for the FSR2 upscaling, I see absolutely zero increase in terms of performance.
.15 driver is out so this is outdated.
proton 8.16x2 2023-10proton 8.0-4x1 2023-10
Starfield, previously kind of fixed with the release of 535.43.10 Vulkan dev driver, is broken again.
This time I've installed 535.43.11 Vulkan dev released yesterday, in order to check whether DGCC improvements mentioned in the NVIDIA's release log had any effect on the performance. Instead, the game hangs again, just like it did before 535.43.10 release. The only difference is that this time I can actually send KILL to the process, without resorting to soft-resetting the whole system.
The behavior is consistent: I've tried running the same Wine version and the same vkd3d-proton version on 535.43.10, and had no issues with the game crashing.
With that said, there's at least two reports from Pascal and Turing users whom are stating that game runs fine for them on 535.43.11 (see https://forums.developer.nvidia.com/t/starfield-does-not-launch-on-most-nvidia-gpus-on-linux-and-not-at-all-on-pascal-gpus/265483/115 and below). Other users, in turn, confirm that 535.43.11 broke things for them.
Can it be clarified whether this is an issue on NV or vkd3d-proton's side?
Software information
Starfield v1.7.33
System information
Log files
starfield_817-tkg_vkd3d_be.txt
Additional logs from another NV user running RTX 3070: https://forums.developer.nvidia.com/t/starfield-does-not-launch-on-most-nvidia-gpus-on-linux-and-not-at-all-on-pascal-gpus/265483/117