A tool to get GPU information from Intel and Apple silicon based macs

by iOS Example

macos-telegraf-gpumonitoring

Sharing a tool to get GPU information from Intel and Apple silicon based macs. This command line tool is used with telegraf to populate an Influxdb bucket to display stats in grafana;

You like it? You use it? It saved you a couple of hours? Hey ??? You can now buy me a coffee! ☕️

Disclaimer, this code has been tested on:

  • Intel Mac Pro with AMD Radeon Pro W6600
  • Apple Silicon Macbook M1 Ultra

It hasn’t been tested on multi-gpu setup but should be quite easily adapted.

Installation process

  • Copy gpuPerformanceStatistics to /usr/local/bin
  • Update your telegraf.conf accordingly (see example bellow)
  • Restart telegraf
  • Update and use provided grafana dashboard json as template
  • Enjoy

GPU performance statistics are exported using Apple IOKitframework as a JSON.

Telegraf is configured to invoke gpuPerformanceStatistics and parse the JSON output to ingest all available fields and send them to influxdb.

Here’s an example of telegraf.conf

[[inputs.execd]]
  command = ["/usr/local/bin/gpuPerformanceStatistics", "IOAccelerator"]
  name_override = "macosgpu"
  signal = "none"
  restart_delay = "10s"
  data_format="json"

[[outputs.influxdb_v2]]
  urls = ["http://yourinfluxdbserver:8086"]
  ## Token for authentication.
  token = "yourtoken"
  
  ## Organization is the name of the organization you wish to write to; must exist.
  organization = "yourorganisation""

  ## Destination bucket to write into.
  ## Change it and update the dashboard accordingly (hint: find/replace into the grafana dashboard json BEFORE importing it into grafana, you'll save time)
  bucket = "XRRMonitoring"

gpuPerformanceStatistics usage

> gpuPerformanceStatistics --help

Usage: gpuPerformanceStatistics className

On Intel and Apple silicon macs you can use the classname IOAccelerator to be safe or you can be more specific with something like AMDRadeonX6000_AMDNavi23GraphicsAccelerator (if you have multiple GPUs. ioreg -l is your friend to find your targets…)

Main AMD GPU metrics (and more)

  • Core Clock (MHz)
  • Memory Clock (MHz)
  • Total Power (W)
  • Device Utilization (%)
  • Temperature (C)
  • Fan Speed (%)
  • Fan Speed (RPM)
  • GPU Activity (%)

Apple Silicon metrics (that’s all)

  • Device Utilization
  • Renderer Utilization
  • Tiler Utilization
  • Alloc system memory
  • In use system memory

You like it? You use it? It saved you a couple of hours? Hey ??? You can now buy me a coffee! ☕️

GitHub

View Github