Related to #3053, which was closed as fixed via NVIDIA driver branch 595/596 ("Vulkan beta drivers 596 improve that a lot"). Reproducing the same Xid 109 hang on a driver branch that should already include that fix, so filing separately rather than commenting on a closed/completed issue.

Hardware/OS:

• GPU: RTX 4070 Super (Ada Lovelace)
• Distro: CachyOS, kernel 7.1.4-1-cachyos (also reproduced on 6.18.38-2-cachyos-lts)
• Drivers tested: 610.43.03 (open kernel modules) and 580.173.02 (proprietary/closed DKMS) — identical hang on both. Note 610.x is numerically after the 595/596 branch that reportedly fixed #3053, so either this is a regression or a distinct-but-identical-looking bug.

Proton builds tested (all hang identically):

• GE-Proton11-1
• Proton Experimental
• proton-cachyos

Symptom: Deterministic freeze right as the post-loading-screen benchmark starts. Confirmed via kernel log every time:

NVRM: Xid (PCI:0000:01:00): 109, ... name=forzahorizon6.e, channel 0x0000XX, errorString CTX SWITCH TIMEOUT, Info 0xXXXXXX

(channel and Info sub-code vary per attempt — reproduced across 8+ separate launches)

Ruled out:

• Ray tracing off (RTReflectionQuality=0, RTGIQuality=0)
• DLSS/DLSSG/FSR3/XeSS all off, tried both DLAA and TAA
• NVIDIATech/Reflex fully disabled
• PCIe ASPM disabled (pcie_aspm=off, confirmed via dmesg)
• PROTON_NO_ESYNC=1 PROTON_NO_FSYNC=1
• VKD3D_CONFIG=no_upload_hvv,single_queue + full pipeline/shader cache wipe (fozpipelinesv6, DXVK_state_cache, nvidiav1)
• Generic GPU instability: vkmark sustained load runs clean, no Xid, confirming this isn't a general hardware/driver fault

Happy to grab any further logs/traces if useful for debugging.