Minimum GPU and VRAM clock adjustments not working

[leonardohn]

Checklist

• The problem is not listed in the hardware support matrix as a known limitation. Examples of known driver-side issues are fans not turning on with a custom curve or inaccurate power limits on RDNA3.
• I've included a debug snapshot if the issue is related to hardware configuration

Bug description

I encountered two issues related to minimum GPU and VRAM clock adjustments while using LACT:

• When changing either the minimum GPU clock or the VRAM clock through the LACT GUI, the GPU experiences severe instability. The GPU fan immediately ramps up to maximum speed. Video output gets glitchy, exactly as described in Feature request: Minimum GPU clock #129.

• When I attempt to set a new minimum GPU or VRAM clock limit by manually editing the configuration file, the daemon appears to ignore the new limit. After restarting the daemon, the configured limits are not applied, and the clocks remain unchanged.

Debug Snapshot: LACT-sysfs-snapshot-20241015-234520.tar.gz

System info

- LACT version: 0.5.6-release
- GPU model: AMD Radeon RX 6900 XT (0x1002:0x73BF:0xC0)
- Kernel version: 6.11.2-cachyos
- Distribution: NixOS 24.11

[leonardohn]

Additional information: If I set the minimum GPU clock using /sys interface, lact-gui displays wrong power states:

[ilya-zlobintsev]

Setting minimum clocks can indeed be iffy, though I have not seen this happen before with GPU clock, only with VRAM.

A few more questions:

• After restarting the daemon, the configured limits are not applied

  Do you mean that the values are not shown at all (in the UI, or manually in the pp_od_clk_voltage file in /sys), or that they are written there but are not the clocks the GPU actually runs at?

• Do the clocks change correctly when you write them manually in /sys? (ignoring the reporting)

• Please include a debug snapshot for when the power state reporting is wrong

• Try setting disable_clocks_cleanup to true in the daemon config, then reboot. The problem might be related to the GPU not properly handling a clocks reset command, which by default is sent every time settings are applied.

[leonardohn]

Setting minimum clocks can indeed be iffy, though I have not seen this happen before with GPU clock, only with VRAM.

I believe the GUI changes both values when you enable the advanced configuration, doesn't it? As the VRAM slider will also be available.

  Do you mean that the values are not shown at all (in the UI, or manually in the `pp_od_clk_voltage` file in `/sys`), or that they are written there but are not the clocks the GPU actually runs at?

The values doesn't seem to be applied anywhere, it just stays in the default minimum.

* Do the clocks change correctly when you write them manually in /sys? (ignoring the reporting)

Yes, but they are displayed incorrectly as in the screenshot.

* Please include a debug snapshot for when the power state reporting is wrong

Here it is: LACT-sysfs-snapshot-20241016-050352.tar.gz

* Try setting `disable_clocks_cleanup` to true in the daemon config, then reboot. The problem might be related to the GPU not properly handling a clocks reset command, which by default is sent every time settings are applied.

I will try and report you back.

[leonardohn]

The problem persists with disable_clocks_cleanup: true.

[ilya-zlobintsev]

I've just tested this on a Sapphire 6900 XT and unfortunately I wasn't able to reproduce any of the issues you're having. Editing both the GPU and VRAM min clock worked correctly via GUI, config file or manually via /sys.