tuned-amdgpu

Hacky solution to integrate AMDGPU power/clock control into tuned profiles with Ansible.

Provides three variations of the tuned profiles found in /usr/lib/tuned using the AMDGPU hwmon interfaces:

• default: the out-of-the-box configuration
• overclock: the optimized card configuration. Includes all of the clock/voltage/power settings
• peak: the same as overclock, but with clock gating removed. May help profiling

Contrary to the name, overclock can be used to de-tune the card as well.

Warning: This is only tested with RX6000 series GPUs, others may not work properly. Use at your own risk!

Profiles

Two 'profiles' are in each name:

• before amdgpu is the source profile provided with tuned
• after amdgpu tells the GPU clock profile offered, outlined below

Output profile	Description
balanced-amdgpu-default	Includes the (assumed) existing balanced tuned profile. Only adjusts the GPU power limit (typically lower). Clocks/voltage curve remain the default.
desktop-amdgpu-overclock	Includes the (assumed) existing desktop tuned profile. Adjusts the GPU power limit, clocks, and the voltage curve.
desktop-amdgpu-peak	Includes the (assumed) existing desktop tuned profile. Same as the overclock profile, but locks clocks to their highest configured values

Config

The playbook will render/make effective this config file: /etc/tuned/amdgpu-profile-vars.conf

Here is a preview:

gpu_clock_min=500
gpu_clock_max=2715
gpumem_clock_static=1075
gpu_power_multi_def=0.869969040247678
gpu_power_multi_oc=1.0
gpu_mv_offset=+60

Changes outside of Ansible are not immediately effective. Switching tuned profiles or restarting the service would be required.

One can use gamemode for dynamic switching. Sample ~/.config/gamemode.ini below:

[custom]
start=tuned-adm profile latency-performance-amdgpu-overclock
end=tuned-adm profile latency-performance-amdgpu-default

See this Arch Wiki link for more comprehensive information.

Variables

These are the variables you'll want to change/consider.

Variable	Description
gpu_clock_min	Sets the min (dynamic) GPU clock (in Mhz) for the non-default amdgpu profiles
gpu_clock_max	Sets the max (dynamic) GPU clock (in MHz) for the non-default amdgpu profiles
gpumem_clock_static	Sets the static memory clock for the GPU (in MHz). This is not the effective data rate. That would be a multiple of this depending on the type of VRAM. To avoid flickering this is not allowed to change with load, only between default and overclock/peak profiles.
gpu_mv_offset	GPU core voltage offset. Takes +/- some integer in millivolts. Can be used to both over and under volt. eg: -50 (undervolt 50mV or 0.05V)
gpu_power_multi_def	Float between 0.0 and 1.0; controls power limit relative to the board capability. For 'default'-named power profiles.
gpu_power_multi_oc	Similar to gpu_power_multi_def, for 'overclock'-named power profiles.