Hello,
I am encountering a reproducible GPU power-state issue on my ROG Strix SCAR 16 (2025) G635LX with an RTX 5090 and BIOS 327. This behavior appears in both standard GPU mode and Eco mode, but in different ways. All observations are based strictly on system measurements and actual outputs.
In standard GPU mode, the system does not enter a proper idle state. With Silent mode active, no monitoring tools running, no background GPU load, and the internal display set to either 60 Hz or 240 Hz, the GPU remains stuck in higher power states. Using nvidia-smi, the memory clock consistently stays between 11001–14001 MHz, the GPU stays in P2 or P0, and the system draws 23–30 watts at idle. Expected behavior would be P8 with a 405 MHz memory clock and single-digit wattage, but this never happens. This behavior is identical across all backlight modes (One-Zone, Multi-Zone, Multi-Zone Strong).
When switching to Eco mode, the discrete NVIDIA GPU is fully powered off. In this state, nvidia-smi reports that it cannot communicate with the NVIDIA driver, which is expected when the GPU is not active. Because the GPU is powered down, idle clock speeds or wattage cannot be measured. However, based on system behavior – specifically that the fans remain silent under conditions where normal mode draws 23–30 watts – it is clear that the system is not consuming the same high idle power in Eco mode.
Despite this, Eco mode introduces a second problem. After restarting the system or waking from modern standby, the laptop consistently triggers a VIDEO_DXGKRNL_FATAL_ERROR (0x113). The minidump identifies the subtype UNEXPECTED_DEFERRED_DESTRUCTION (0x19), with the crash occurring in dxgkrnl!DpiFdoHandleDevicePower, meaning the failure happens while the system attempts to power the discrete GPU back on. The NVIDIA module nvlddmkm.sys stops responding during this transition. Hyper-V is active on the system, but the crash is clearly tied to the GPU power-on sequence itself. I tested this again by restarting in Eco mode, and the same BSOD occurred immediately.
In summary, standard GPU mode results in permanently elevated idle clocks and continuous 23–30 watt idle consumption, while Eco mode shuts down the GPU correctly but causes a reproducible crash whenever the GPU needs to be powered back on. BIOS 327 does not resolve either issue. The symptoms point toward a firmware-level GPU power-state sequencing or ACPI handling problem.
I would like to request that this be forwarded to the firmware and ACPI engineering teams. If a beta BIOS is available, I am willing to test it. I can also provide ETW ACPI traces, minidumps, and other logs upon request.
Additionally, I would like to ask whether any other users with the same hardware configuration (G635/G635LX with RTX 5090) are experiencing similar behavior. If anyone can check their idle GPU wattage in standard mode using nvidia-smi and report whether they also see the GPU stuck in P2/P0 with elevated memory clocks, it would help determine whether this issue is isolated or widespread.
Thank you.
