Adjust power management settings for Ryzen Mobile Processors.
Based on: FlyGoat/ryzen_nb_smu
RyzenAdjUI_WPF by "JustSkill" is no longer maintained, for GUI please see Universal x86 Tuning Utility or ryzen-controller-team/ryzen-controller.
The command line interface is identical on both Windows and Unix-Like OS.
You should run it with Administrator on Windows or root on Linux.
You can write a shell script or bat to do it automatically.
$./ryzenadj -h
Usage: ryzenadj [options]
Ryzen Power Management adjust tool.
-h, --help show this help message and exit
Options
-i, --info Show information and most important power metrics after adjustment
--dump-table Show whole power metric table before and after adjustment
Settings
-a, --stapm-limit=<u32> Sustained Power Limit - STAPM LIMIT (mW)
-b, --fast-limit=<u32> Actual Power Limit - PPT LIMIT FAST (mW)
-c, --slow-limit=<u32> Average Power Limit - PPT LIMIT SLOW (mW)
-d, --slow-time=<u32> Slow PPT Constant Time (s)
-e, --stapm-time=<u32> STAPM constant time (s)
-f, --tctl-temp=<u32> Tctl Temperature Limit (degree C)
-g, --vrm-current=<u32> VRM Current Limit - TDC LIMIT VDD (mA)
-j, --vrmsoc-current=<u32> VRM SoC Current Limit - TDC LIMIT SoC (mA)
-k, --vrmmax-current=<u32> VRM Maximum Current Limit - EDC LIMIT VDD (mA)
-l, --vrmsocmax-current=<u32> VRM SoC Maximum Current Limit - EDC LIMIT SoC (mA)
-m, --psi0-current=<u32> PSI0 VDD Current Limit (mA)
-n, --psi0soc-current=<u32> PSI0 SoC Current Limit (mA)
-o, --max-socclk-frequency=<u32> Maximum SoC Clock Frequency (MHz)
-p, --min-socclk-frequency=<u32> Minimum SoC Clock Frequency (MHz)
-q, --max-fclk-frequency=<u32> Maximum Transmission (CPU-GPU) Frequency (MHz)
-r, --min-fclk-frequency=<u32> Minimum Transmission (CPU-GPU) Frequency (MHz)
-s, --max-vcn=<u32> Maximum Video Core Next (VCE - Video Coding Engine) (MHz)
-t, --min-vcn=<u32> Minimum Video Core Next (VCE - Video Coding Engine) (MHz)
-u, --max-lclk=<u32> Maximum Data Launch Clock (MHz)
-v, --min-lclk=<u32> Minimum Data Launch Clock (MHz)
-w, --max-gfxclk=<u32> Maximum GFX Clock (MHz)
-x, --min-gfxclk=<u32> Minimum GFX Clock (MHz)
-y, --prochot-deassertion-ramp=<u32> Ramp Time After Prochot is Deasserted: limit power based on value, higher values does apply tighter limits after prochot is over
--apu-skin-temp=<u32> APU Skin Temperature Limit - STT LIMIT APU (degree C)
--dgpu-skin-temp=<u32> dGPU Skin Temperature Limit - STT LIMIT dGPU (degree C)
--apu-slow-limit=<u32> APU PPT Slow Power limit for A+A dGPU platform - PPT LIMIT APU (mW)
--skin-temp-limit=<u32> Skin Temperature Power Limit (mW)
--power-saving Hidden options to improve power efficiency (is set when AC unplugged): behavior depends on CPU generation, Device and Manufacture
--max-performance Hidden options to improve performance (is set when AC plugged in): behavior depends on CPU generation, Device and Manufacture
If I'm going to set all the Power Limit to 45W, and Tctl to 90 °C, then the command line should be:
./ryzenadj --stapm-limit=45000 --fast-limit=45000 --slow-limit=45000 --tctl-temp=90
You don't need to install RyzenAdj because it does not need configuration, everything is set via arguments However, some settings could get overwritten by power management features of your device, and you need to regularly set your values again.
We did provide some examples for automation. And these require configuration during installation.
Because it is very easy to build the latest version of RyzenAdj on Linux, we don't provide precompiled packages for distributions. Just follow the build instructions below and you are ready to use it.
Before you start installing anything, it is highly recommended getting familiar with RyzenAdj to find out what can be done on your device.
Use the CLI ryzenadj.exe to test the support of your device and to benchmark the effects of each setting.
If your values don't stay persistent you may want to consider installing our example script for automation.
- Prepare your favorite RyzenAdj arguments
- Copy the content of your RyzenAdj folder to the final destination
- Put your configuration into
readjustService.ps1and test it as administrator until everything works as expected - Install
readjustService.ps1as Task for Windows Task Scheduler by runninginstallServiceTask.bat
Deinstallation of the Task can be done via uninstallServiceTask.bat
Over Windows Task Scheduler you can check if it is running. It is called RyzenAdj below AMD folder.
Or just run
SCHTASKS /query /TN "AMD\RyzenAdj"
Building this tool requires C & C++ compilers as well as cmake.
RyzenAdj needs elevated access to the NB config space. This can be achieved by using either one of these two methods:
- Using libpci and exposing
/dev/mem - Using the ryzen_smu kernel module
RyzenAdj will try ryzen_smu first, and then fallback to /dev/mem, if no compatible smu driver is found.
The minimum supported version of ryzen_smu is 0.1.7
If no backend is available, RyzenAdj will fail initialization.
Please note that /dev/mem access may be restricted, for security reasons, in your kernel config
Please make sure that you have libpci dependency before compiling.
On Debian-based distros this is covered by installing pcilib-dev package:
sudo apt install build-essential cmake libpci-dev
On Fedora:
sudo dnf install cmake gcc-c++ pciutils-devel
On Arch:
sudo pacman -S base-devel pciutils cmake
On OpenSUSE Tumbleweed:
sudo zypper in cmake gcc14-c++ pciutils-devel
You may need to add the iomem=relaxed param to your kernel params on Tumbleweed, or you may run into errors at runtime.
If your Distribution is not supported, try finding the packages or use Distrobox or Toolbox instead.
The simplest way to build it:
git clone https://github.com/FlyGoat/RyzenAdj.git
cd RyzenAdj
rm -r win32
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make
if [ -d ~/.local/bin ]; then ln -s $(readlink -f ryzenadj) ~/.local/bin/ryzenadj && echo "symlinked to ~/.local/bin/ryzenadj"; fi
if [ -d ~/.bin ]; then ln -s $(readlink -f ryzenadj) ~/.bin/ryzenadj && echo "symlinked to ~/.bin/ryzenadj"; fi
To let RyzenAdj use ryzen_smu module, you have to install it first, it is not part of the linux kernel.
On Fedora:
sudo dnf install cmake gcc gcc-c++ dkms opensslClone and install ryzen_smu:
git clone https://github.com/amkillam/ryzen_smu # Active fork of the original module
(cd ryzen_smu/ && sudo make dkms-install)If you are using secure boot, you have to enroll the UEFI keys which dkms has generated on its first run. These have to be added to your machines UEFI key database. This can be done with following command, which will ask you to set a password. This password is only needed one single time later in the MOK manager.
sudo mokutil --import /var/lib/dkms/mok.pubRestart your system. This will boot into the MOK manager. Choose Enroll MOK, enter your password
and then reboot.
Here
are some screenshots describing the process.
The module is now loaded and visible via dmesg. It will show a message about the kernel being tainted, but this just means it loaded a (potentially proprietary) binary blob.
Build and install RyzenAdj:
git clone https://github.com/FlyGoat/RyzenAdj
cd RyzenAdj
cmake -B build -DCMAKE_BUILD_TYPE=Release
make -C build -j"$(nproc)"
sudo cp -v build/ryzenadj /usr/local/bin/It can be built by Visual Studio + MSVC automaticaly, or Clang + Nmake in command line. However, as for now, MingW-gcc can't be used to compile for some reason.
Required dll is included in ./win32 of source tree. Please put the dll library and sys driver in the same folder with ryzenadj.exe.
We don't recommend you to build by yourself on Windows since the environment configuarion is very complicated. If you would like to use ryzenadj functions in your program, see libryzenadj.