[Support]: Intel statistics - unraid way?

[ProsjektX]

With the known and returning issue of intel (i)gpu statistics not updating after a period after power-up, would it be possible to implement this in the way that the unraid "gpu statistics" plugin is doing this, which always work and correctly reports loads, memory usage etc?

https://github.com/b3rs3rk/gpustat-unraid/blob/dev/src/gpustat/usr/local/emhttp/plugins/gpustat/lib/Intel.php

[NickM-27]

It uses the exact same utility that Frigate does

https://github.com/b3rs3rk/gpustat-unraid/blob/ddaac43afb038a88e3688007b4746de09ed0fc64/src/gpustat/usr/local/emhttp/plugins/gpustat/lib/Intel.php#L37

[ProsjektX]

Yes, I saw that, but it seems it get's executed once for each sample/update, running only for 250ms, and get's terminated if running more than 1.3 seconds, so not running continuously in the background. Perhaps that's the way around the issue?

[NickM-27]

that is very similar to how Frigate does it, it does not run it continuously

[ProsjektX]

I took a quick look at this, and the major issue with frigate seems to be you use the "engines" field of the json output, which should contain the accumulated total load, but it does not, or is unreliable. Testing running intel_gpu_top directly under unraid, and in the container. Both report 0%, but do report for the individual load for the "clients" (processes), which is what the unraid plugin uses to sum up the load (for compute, video, videoenhance,...). Perhaps you could implement it the same way, to work around the reported total not being reliable, as this is a returning confusion.

[NickM-27]

The clients list is not always present, it depends on your driver version and what hardware you have. This is just something that Intel needs to fix