RTX 5090 Xid 109 CTX SWITCH TIMEOUT crash on Linux 595.58.03 — Crimson Desert

Graphics / Linux Linux
1

GPU: RTX 5090 (Blackwell)
Driver: 595.58.03 (nvidia-open)
Kernel: 6.19.10-zen1-1-zen (linux-zen)
Distro: Arch Linux

Crimson Desert (Steam App ID 3321460, via Proton) crashes immediately on
loading a save or starting a new game. The GPU resets and the game hangs.

Relevant kernel log (journalctl -k -b -1):

NVRM: GPU at PCI:0000:0b:00: GPU-47ac047a-5039-307f-00cb-212eba63fc08
NVRM: Xid (PCI:0000:0b:00): 109, pid=6401, name=CrimsonDesert.e, channel 0x00000014,

errorString CTX SWITCH TIMEOUT, Info 0x38016
NVRM: Xid (PCI:0000:0b:00): 109, pid=20829, name=CrimsonDesert.e, channel 0x00000014,
errorString CTX SWITCH TIMEOUT, Info 0x3c016

The crash is reproducible — happens on both save load and new game start.

Troubleshooting already attempted:

  • Verified game files, cleared shader cache
  • Tested Proton Experimental and GE-Proton10-34
  • VKD3D_CONFIG=no_upload_hvv,nodxr,single_queue
  • Wiped Proton prefix
  • nvidia.NVreg_EnableResizableBar=1 (no effect)
  • nvidia.NVreg_EnableGpuFirmware=0 (ignored on Blackwell — GSP mandatory)
  • vkd3d-proton-mingw-git from AUR with PROTON_USE_SYSTEM_VKD3D=1

Other Proton games (e.g. Kingdom Come Deliverance 2) work fine on the same
system, suggesting this is specific to Crimson Desert’s GPU workload
triggering the Xid 109 on Blackwell.

Is this a known issue with 595.x on Blackwell? Is there a workaround or
planned fix in an upcoming driver release?

nvidia-bug-report.log.gz (705.8 KB)

2 Likes
2

Hi soundofmind,

I went through your nvidia-bug-report.log. Here’s what the data shows.

Three crashes, same channel, across two separate boots

From your journalctl sections in the bug report:

Boot -2 (NVreg_EnableResizableBar=1):
  06:51:32 — Xid 109, pid=6855,  channel 0x00000014, Info 0x38016

Boot -1 (NVreg_EnableGpuFirmware=0):
  07:04:49 — Xid 109, pid=6401,  channel 0x00000014, Info 0x38016
  07:13:19 — Xid 109, pid=20829, channel 0x00000014, Info 0x3c016

Channel 0x00000014 is allocated every time the game initializes regardless of what kernel parameters you boot with. Same channel across reboots with different parameters rules out transient state — the crash is deterministic and structural.

NVreg_EnableGpuFirmware=0 is silently ignored on GB202

Your boot -1 cmdline has nvidia.NVreg_EnableGpuFirmware=0. Your bug report header shows GSP firmware version 98.02.2E.00.D4 loaded and active regardless. On Blackwell (GB202), GSP is mandatory — the flag is accepted by the kernel module but has no effect. The standard pre-Blackwell workaround for Xid 109 (falling back to RM-managed context scheduling) is not available on your hardware.

No precursor warnings before the Xid fires

The kernel log between drm init and the first Xid 109 is completely clean — no GSP RPC timeout, no PBDMA fault, no memory error. The watchdog fires silently with no logged precursor.

PCIe link negotiated at Gen4 x16, not Gen5

From your log:

0b:00.0: 252.048 Gb/s available, limited by 16.0 GT/s PCIe x16
          capable of 504.112 Gb/s with 32.0 GT/s PCIe x16

GB202 is rated for PCIe 5.0 x16. Your slot is running at PCIe 4.0 x16. Platform/BIOS issue, not a driver bug, but worth noting.

MSI-X interrupt vector distribution

From /proc/interrupts in your report:

IRQ 191:  21,636 counts — nvidia MSI-X vector 0
IRQ 192:       0 counts — nvidia MSI-X vector 1
IRQ 193: 167,171 counts — nvidia MSI-X vector 2
IRQs 194–198:   0 counts — nvidia MSI-X vectors 3–7

6 of 8 allocated MSI-X vectors are taking zero interrupts. Flagging for NVIDIA engineers.

Post-crash G-SYNC failures

After the Xid 109 events your log shows:

07:20:40 — nvidia-modeset: WARNING: GPU:0: ROG PG279Q (DP-2): Failed to initialize G-SYNC
07:21:40 — nvidia-modeset: WARNING: GPU:0: ROG PG279Q (DP-2): Failed to initialize G-SYNC

GPU recovered enough to keep the display alive but the display engine did not fully reinitialize after the reset.

NVIDIA has this filed internally

This exact issue — Crimson Desert softlock on Blackwell with 595.58.03 — is confirmed Blackwell-only by multiple reporters and is tracked internally by NVIDIA as bug 6017526, confirmed by NVIDIA staff in the 595 release feedback thread:
595 release feedback & discussion (posts 227–228)

A second RTX 5090 reporter (seldomridgej) confirmed the identical symptom — freeze on loading screen after pressing play or new game, Proton version irrelevant:
595 release feedback & discussion (post 209)

3

Use ProtonUp-Qt to download proton-cachyos and then launch the game with this PROTON_VKD3D_HEAP=1 %command%
It’s a known issue with the latest drivers 595.x

3 Likes
4

GPU: RTX 5080 (Blackwell)
Driver: 595.58.03 (nvidia-open)
Kernel: Linux 6.19.12-1-cachyos
Distro: CachyOS

Proton version: Proton CachyOS 10.0-20260407

I got the same issue. the parameters did not work for me.

5

Paradise Nowhere on Steam this free demo freezes every time after starting a new game with NVRM: Xid (PCI:0000:01:00): 109, pid=4070210, name=GameThread, channel 0x00000033, errorString CTX SWITCH TIMEOUT, Info 0x8c023

GPU: RTX 5090
Driver: 595.58.03
Kernel: 6.19.11
Distro: Debian testing