Wiki

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Ubuntu - GPU - AMD GPU - Overclocking - Increase Power on AMD GPUs ======

It is possible to increase the number of watts supplied to the GPU.

  * This should increase performance.


<WRAP alert>
**ALERT:**  This will effectively overclock the GPU.

  * This may result in erratic behavior being seen on the system (i.e. frequent crashing).
  * If so, restore the GRUB file back to what it was before!

</WRAP>

<WRAP info>
**NOTE:**  It is a best practice to increase voltage in small increments and then test for stability.

  * This can be done by running your 3D games/applications at your preferred settings.
  * If the adjustments made above causes stability issues, then restore factory default settings.

</WRAP>

<WRAP info>
**NOTE:**  Voltages are not shared between GPU and Memory Clocks; both are set independently.
</WRAP>

----

===== Edit GRUB Config =====

==== Make a Backup of GRUB ====

<code bash>
sudo cp /etc/default/grub /etc/default/grub.bak
</code>

<WRAP info>
**NOTE:**  If the system somehow breaks, this can be used to restore the backup file.
</WRAP>

----

==== Edit GRUB Configuration ====

Edit **/etc/default/grub**.

Change:

<file bash /etc/default/grub>
...
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash"
...
</file>

to

<file bash /etc/default/grub>
...
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash amdgpu.ppfeaturemask=0xffffffff"
...
</file>

<WRAP info>
**NOTE:**  Appending this parameter will give “full control” of the GPU.

  * This will allow more advanced options to be chosen for Voltage, GPU min/max frequency, and Memory speed.

</WRAP>

----

==== Regenerate Bootloader Configuration ====

<code bash>
sudo update-grub

or

sudo grub-mkconfig -o /boot/grub/grub.cfg
</code>

<WRAP info>
**NOTE:**  On a Fedora or openSUSE distro use:

<code bash>
sudo grub2-mkconfig -o /boot/grub2/grub.cfg
</code>

</WRAP>

----

==== Reboot ====

After rebooting, check the Changes.

Verify the current featuremask after reboot by checking here:

<code bash>
printf "0x%08x\n" $(cat /sys/module/amdgpu/parameters/ppfeaturemask)
</code>

----

===== Check the current performance levels =====

<code bash>
cat /sys/class/drm/card0/device/power_dpm_force_performance_level
</code>

returns:

<code bash>
auto
</code>

<WRAP info>
**NOTE:**  To allow changing the performance, this has to be set to **manual** control.

  * If not, then the following error will be seen: **Invalid argument errors when writing clock values of you don't.**

</WRAP>

----

===== Set manual control =====

<code bash>
echo "manual" > /sys/class/drm/card0/device/power_dpm_force_performance_level
</code>

<WRAP info>
**NOTE:**  Other available **power_dpm_force_performance_level** settings are:

  * **auto**: Drivers chooses automatically.
  * **low**: Forces the lowest possible clock and locks the GPU there.
  * **high**: Forces the highest possible clock and locks the GPU there.
  * **profile_standard**, **profile_min_sclk**, **profile_min_mclk**, **profile_peak**: When the __profiling modes__ are selected, clock and power gating are disabled and the clocks are set for different profiling cases. 
    * This mode is recommended for profiling specific work loads where you do not want clock or power gating for clock fluctuation to interfere with your results.
    * **profile_standard** sets the clocks to a fixed clock level which varies from asic to asic.
    * **profile_min_sclk** forces the sclk to the lowest level. 
    * **profile_min_mclk** forces the mclk to the lowest level.
    * **profile_peak** sets all clocks (mclk, sclk, pcie) to the highest levels.
</WRAP>

----

===== Find the GPU device location =====

<code bash>
readlink -f /sys/class/drm/card0/device
</code>

returns:

<code bash>
/sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0
</code>

----

===== Set the Voltage =====

<WRAP info>
**NOTE:**  P-states (Power States) are levels, L0 to L7. that indicate the frequency and voltage of the GPU processor at that state.

  * P0 is the lowest, when the system is idle, as normally do not need higher voltage and frequency.
  * P6/P7 are what are used when gaming,
  * P7 is the highest, when very heavy GPU usage is used.
    * But note that P7 uses more voltage and higher frequency; which results in more heat.
    * Often P7 is not needed.

</WRAP>

To edit the P-state #7 for the core and #2 for the VRAM, as those are the values that our GPU is going to run at while under load.

Edit **/sys/class/drm/card0/device/pp_od_clk_voltage**:

<file bash /sys/class/drm/card0/device/pp_od_clk_voltage>
sudo sh -c "echo 's 7 1450 1150' > /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage"
sudo sh -c "echo 'm 2 2065 950' > /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage"
</file>

<WRAP info>
**NOTE:**  Ensure that the file path is correct per the previous command to discover the path.

  * The values for the clock speeds and voltages will also be different for each GPU.
  * To undervolt, change each P-State voltage for core and VRAM to be less than the default values.
  * The clock speeds might be able to be kept the same, but in some cases this will also have to be reduced.
    * This is where stability testing and benchmarking the GPU comes into play/
</WRAP>

----

===== Apply the changes =====

<WRAP alert>
**WARNING:** Double check the entered values, as mistakes might instantly cause fatal hardware damage!
</WRAP>

<code bash>
echo "c" > /sys/class/drm/card0/device/pp_od_clk_voltage
</code>

To check if it worked out, read out clocks and voltage under 3D load:

<code bash>
sudo watch -n 0.5  cat /sys/kernel/debug/dri/0/amdgpu_pm_info
</code>


<WRAP info>
**NOTE:**  The default values can be reset:

<code bash>
echo "r" > /sys/class/drm/card0/device/pp_od_clk_voltage
</code>
</WRAP>

----

===== Forbid the driver to switch to certain P-states =====

This can be used to workaround problems with deep powersaving P-states like flickering artifacts or stutter.

To force the highest VRAM P-state on a Polaris RX 5xx card, while still allowing the GPU itself to run with lower clocks:

<code bash>
echo "manual" > /sys/class/drm/card0/device/power_dpm_force_performance_level
echo "2" >  /sys/class/drm/card0/device/pp_dpm_mclk
</code>

Allow only the three highest GPU P-states:

<code bash>
echo "5 6 7" >  /sys/class/drm/card0/device/pp_dpm_sclk
</code>

To set the allowed maximum power consumption of the GPU to e.g. 50 Watts:

<code bash>
echo 50000000 > /sys/class/drm/card0/device/hwmon/hwmon0/power1_cap
</code>

----

===== Across boots =====

To enable these changes to carry over across boots create a script and systemd service.

  * Use the values that have been tested to work and the file path to the GPU.

----

==== Create a script ====

<code bash>
#!/bin/sh
sudo sh -c "echo 's 7 <Your desired values here>' > /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage"
sudo sh -c "echo 'm 2 <Your desired values here' > /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage"
sudo sh -c "echo 'c' > /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage"
</code>

  * Save this script in **/usr/bin/** or a safe place for systemwide scripts.


Create a file in **/etc/systemd/system/** like **undervolt.service**:

<file bash undervolt.service>
[Unit]
Description=Undervolting GPU

[Service]
Type=oneshot
ExecStart=/Path/To/Script.sh

[Install]
WantedBy=multi-user.target
</file>

----

==== Enable the service ====

<code bash>
sudo systemctl enable undervolt.service
</code>

----

==== Reboot ====

----

==== Verify ====

<code bash>
cat /sys/devices/pci0000:00/0000:00:03.1/0000:0a:00.0/0000:0b:00.0/0000:0c:00.0/pp_od_clk_voltage
</code>

<WRAP info>
**NOTE:**  It is recommended to make a similar script to revert things to default values.
</WRAP>

----