amdgpu_top

Crates.ioamdgpu_top
lib.rsamdgpu_top
version0.10.0
sourcesrc
created_at2023-05-02 12:06:23.553579
updated_at2024-11-28 09:16:28.171988
descriptionTool to displays AMDGPU usage. The tool displays information gathered from performance counters (GRBM, GRBM2), sensors, fdinfo, gpu_metrics and AMDGPU driver.
homepage
repositoryhttps://github.com/Umio-Yasuno/amdgpu_top
max_upload_size
id854492
size171,177
Umio Yasuno (Umio-Yasuno)

documentation

README

AMDGPU_TOP

amdgpu_top is tool that display AMD GPU utilization, like umr or clbr/radeontop or intel_gpu_top.
The tool displays information gathered from performance counters (GRBM, GRBM2), sensors, fdinfo, and AMDGPU driver.

Simple TUI
(like nvidia-smi, rocm-smi)
TUI GUI
amdgpu_top Simple TUI amdgpu_top TUI amdgpu_top GUI mode

Quick links

Dependent dynamic libraries

  • libdrm
  • libdrm_amdgpu

Usage

cargo run -- [options ..]
# or
amdgpu_top [options ..]
# Run TUI mode
amdgpu_top

# Run GUI mode
amdgpu_top --gui

# Run SMI mode
amdgpu_top --smi

# Dump AMDGPU info
amdgpu_top -d

# Dump AMDGPU info and gpu_metrics
amdgpu_top -d -gm

# Dump AMDGPU info and pp_table
amdgpu_top -d --pp-table

# Dump AMDGPU info in JSON format
amdgpu_top -d --json

# Dump DRM info
amdgpu_top --drm-info

# Decode gpu_metrics file
amdgpu_top --decode-gm <path>

# Decode gpu_metrics file and output in JSON format
amdgpu_top --json --decode-gm <path>

Options

FLAGS:
   -d, --dump
       Dump AMDGPU info. (Specifications, VRAM, PCI, ResizableBAR, VBIOS, Video caps)
       This option can be combined with the "-J" option.
   --list
       Display a list of AMDGPU devices.
   -J, --json
       Output JSON formatted data.
       This option can be combined with the "-d" option.
   --gui
       Launch GUI mode.
   --smi
       Launch Simple TUI mode. (like nvidia-smi, rocm-smi)
   -p, --process
       Dump All GPU processes and memory usage per process.
   --apu, --select-apu
       Select APU instance.
   --single, --single-gpu
       Display only the selected APU/GPU
   --no-pc
       The application does not read the performance counter (GRBM, GRBM2)
       if this flag is set.
       Reading the performance counter may deactivate the power saving feature of APU/GPU.
   -gm, --gpu_metrics, --gpu-metrics
       Dump gpu_metrics for all AMD GPUs.
       https://www.kernel.org/doc/html/latest/gpu/amdgpu/thermal.html#gpu-metrics
   --pp_table, --pp-table
       Dump pp_table from sysfs and VBIOS for all AMD GPUs.
       (only support Navi1x and Navi2x, Navi3x)
   --drm_info, --drm-info
       Dump DRM info.
       Inspired by https://gitlab.freedesktop.org/emersion/drm_info
   --dark, --dark-mode
       Set to the dark mode. (TUI/GUI)
   --light, --light-mode
       Set to the light mode. (TUI/GUI)
   --gl, --opengl
       Use OpenGL API to the GUI backend.
   --vk, --vulkan
       Use Vulkan API to the GUI backend,
       and use APU/iGPU for GUI rendering if it is available.
   -V, --version
       Print version information.
   -h, --help
       Print help information.

OPTIONS:
   -i <usize>
       Select GPU instance.
   --pci <String>
       Specifying PCI path. (domain:bus:dev.func)
   -s <u64>, -s <u64>ms
       Refresh period (interval) in milliseconds for JSON mode. (default: 1000ms)
   -n <u32>
       Specifies the maximum number of iteration for JSON mode.
       If 0 is specified, it will be an infinite loop. (default: 0)
   -u <u64>, --update-process-index <u64>
       Update interval in seconds of the process index for fdinfo. (default: 5s)
   --json_fifo, --json-fifo <String>
       Output JSON formatted data to FIFO (named pipe) for other application and scripts.
   --decode-gm <Path>, --decode-gpu-metrics <Path>
       Decode the specified gpu_metrics file.

Commands for TUI

key
g toggle GRBM
r toggle GRBM2
v toggle VRAM/GTT Usage
f toggle fdinfo
n toggle Sensors
m toggle GPU Metrics
h change update interval (high = 100ms, low = 1000ms)
q Quit
P sort fdinfo by pid
M sort fdinfo by VRAM usage
G sort fdinfo by GFX usage
M sort fdinfo by MediaEngine usage
R reverse sort

Example of using JSON mode

$ amdgpu_top --json | jq -c -r '(.devices[] |
  (.Info | .DeviceName + " (" + .PCI + "): ") +
  ([.gpu_activity | to_entries[] | .key + ": " + (.value.value|tostring) + .value.unit] |
  join(", ")))'

Output

AMD Radeon RX 6600 (0000:03:00.0): GFX: 13%, MediaEngine: 0%, Memory: 4%
AMD Radeon Graphics (0000:08:00.0): GFX: 0%, MediaEngine: 0%, Memory: null%
AMD Radeon RX 6600 (0000:03:00.0): GFX: 15%, MediaEngine: 0%, Memory: 5%
AMD Radeon Graphics (0000:08:00.0): GFX: 0%, MediaEngine: 0%, Memory: null%
AMD Radeon RX 6600 (0000:03:00.0): GFX: 3%, MediaEngine: 0%, Memory: 2%
AMD Radeon Graphics (0000:08:00.0): GFX: 0%, MediaEngine: 0%, Memory: null%
...

Installation

Packages

Build from source

cargo install amdgpu_top

# or

git clone https://github.com/Umio-Yasuno/amdgpu_top
cd amdgpu_top
cargo install --locked --path .

without GUI

cargo install --locked --path . --no-default-features --features="tui"

Distribution specific instructions

Debian/Ubuntu
sudo apt install libdrm-dev

Binary Size

Features Size (stripped)
json ~1.1M
tui ~1.5M
json, tui ~1.7M
json, tui, gui ~18M

References

Translations

amdgpu_top is using cargo-i18n with Project Fluent for translation.
Please refer to pop-os/popsicle for additional supported languages.

Supported Languages

Alternatives

If amdgpu_top is not enough for you or you don't like it, try the following applications.

Commit count: 821

cargo fmt