[NVIDIA DRM] GPU stuck in low-power state (~130W) after first game exit - requires gamescope restart #2041
Description
Is there an existing issue for this?
- I have searched the existing issues
Are you using any gamescope patches or a forked version of gamescope?
- The issue occurs on upstream gamescope without any modifications
Current Behavior
When running gamescope in embedded/DRM mode (gamescope-session), the first game launch after gamescope starts works perfectly - GPU runs at full power (~350W), smooth performance.
However, after exiting the game and launching any game again (same game or different), the GPU becomes stuck in a low-power state (~130W) despite high GPU utilization (90%+). Games run terribly (feels like 20fps) even though the GPU reports high utilization.
This persists for all subsequent game launches until gamescope is restarted or the system is rebooted.
First launch (works correctly):
clocks.current.memory [MHz], pstate, power.draw [W]
10251 MHz, P2, 350.22 W
10251 MHz, P2, 356.83 W
10251 MHz, P2, 355.75 W
Game runs smoothly at expected framerates.
Second launch (broken):
clocks.current.memory [MHz], pstate, power.draw [W]
10251 MHz, P2, 103.91 W
10251 MHz, P2, 130.96 W
10251 MHz, P2, 132.85 W
Game runs terribly despite 90%+ GPU utilization. Power draw is capped at ~130W instead of 350W+.
Key observations:
- P-state remains P2 in both cases - it's the power draw that's wrong, not the performance state
- Memory clocks are identical (10251 MHz) in both states
- GPU utilization is high (90%+) in both cases, but power is severely limited
- Restarting gamescope fixes the issue until the next game exit/relaunch cycle
- Full system reboot also fixes it
- Issue is not game-specific - happens with any game
- Issue is not Proton-specific - persists across different Proton versions
Expected Behavior
GPU should return to full-power operation when a new game is launched, regardless of whether a previous game was run in the same gamescope session.
Steps To Reproduce
- Boot into gamescope-session (embedded/DRM mode)
- Launch any game from Steam
- Verify game runs well, check nvidia-smi shows ~300-400W power draw
- Exit the game completely (back to Steam UI)
- Launch any game again (same or different)
- Observe severe performance degradation
- Check nvidia-smi - power draw is now stuck at ~130W despite high utilization
Hardware information
- Distro: CachyOS Linux (rolling)
- Kernel: 6.18.1-2-cachyos
- CPU: AMD Ryzen 7 7800X3D 8-Core Processor
- GPU: NVIDIA GeForce RTX 4090
- Driver Version: 580.119.02Software information
- Desktop environment: gamescope-session (Steam Big Picture / embedded mode)
- Session type: DRM (embedded gamescope, no desktop environment)
- Gamescope version: 3.16.17 (gcc 15.2.1)
- Gamescope launch command(s): /usr/bin/gamescope --prefer-output * --xwayland-count 2 --hide-cursor-delay 3000 --fade-out-duration 200 --steamWhich gamescope backends have the issue you are reporting?
- Wayland (default for nested gamescope)
- DRM (default for embedded gamescope, i.e. gamescope-session)
- SDL
- OpenVR
Logging, screenshots, or anything else
nvidia-smi during bad state shows no throttling reasons active:
Clocks Event Reasons
Idle : Not Active
Applications Clocks Setting : Not Active
SW Power Cap : Not Active
HW Slowdown : Not Active
Sync Boost : Not Active
SW Thermal Slowdown : Not Active
Display Clock Setting : Not Active
PCIe link speed remains correct (16.0 GT/s) in both states.
Tested with completely stock gamescope-session config (no custom GAMESCOPECMD) - issue persists.
Attempting to force GPU clocks via nvidia-smi -lgc and nvidia-smi -lmc has no effect while in bad state.
Workaround:
Restart gamescope between game sessions:
systemctl --user restart [email protected]
Could not test nested gamescope - black screen on Nvidia + Wayland.