Differences

This shows you the differences between two versions of the page.

--- ubuntu:gpu:amd_gpu:overclocking:increase_power_on_amd_gpus [2023/06/27 10:04] – created peter
+++ ubuntu:gpu:amd_gpu:overclocking:increase_power_on_amd_gpus [2023/06/27 13:45] (current) – [Set the Voltage] peter
@@ Line 4: / Line 4: @@
   * 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>
 ----
-Append **amdgpu.ppfeaturemask=0xffffffff** to the GRUB configuration file.
+===== Edit GRUB Config =====
-Appending this parameter will also allow us to have “full control” of the GPU; we’ll have more advanced options to choose from in CoreCtrl as far as voltage, GPU min/max frequency, and memory speed.
+==== 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>
-**ALERT**  This will effectively overclock the GPU.
+**WARNING:** Double check the entered values, as mistakes might instantly cause fatal hardware damage!
+</WRAP>
-  * This may result in erratic behavior being seen on the system (i.e. frequent crashing).
+<code bash>
-  * If so, restore the GRUB file back to what it was before!
+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>
+----