nvidia_gpu block resumes GPU from suspend on Optimus devices

[GaspardCulis]

System info

CPU : AMD Ryzen 9 5900HS with Radeon Graphics GPU : NVIDIA GeForce RTX 3070 Mobile iGPU : AMD ATI Radeon Vega Series

Issue

On Laptops with both an iGPU and a dedicated GPU, the dedicated GPU can be suspended to lower power draw, and gets resumed when high performance is needed (PRIME - Arch Wiki).

I can query the GPU state with this command:

cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

I noticed that the nvidia_gpu block resumes my dedicated GPU from its suspended state.

I suggest that a new block option should get added that checks if the GPU is suspended and avoids resuming it.

[MaxVerevkin]

Reported in #1880.

I suggest that a new block option should get added that checks if the GPU is suspended and avoids resuming it.

By "avoids resuming" do you mean stops updating?

[GaspardCulis]

Just displaying something like this instead of waking up the GPU only to get its stats:

Block config

[[block]]
block = "custom"
command = "cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status"
interval = 2
format = " \uf26c   $text "

And when you detect that it's active, just display the stuff normally.

I tried to achieve this behavior with a special toml config but couldn't get it working as intended without just doing the whole thing (check for power state, and then retrieve all of the info) in a script.

[MaxVerevkin]

Ah, okay, this sounds good. Maybe we can add suspended_format to customize the text, or even hide the block.

cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

Do you know a generic way to get this path?

Also, I don't have an nvidia gpu to play with, so would you be interested in implementing this?

[GaspardCulis]

This 000:01:00.0 part of the path refers to the PCI slot of the NVIDIA GPU retrieved with lspci ([domain:]bus:device.function). I'll have to find a way to find this reliably. I have some rust projects but I'm still a beginner. I'll take a look at the code in a few days and tell you if I'm confident in implementing this.

[GaspardCulis]

Took a look into the code, I'm confident, but after putting more toughs into the thing, I don't know what to implement, thing is, if we wait for the GPU to wake up before retrieving the info, the info retrieval process is going to keep it active.

[Ergus]

Hi I was having the same issue. My workaround was to change the retrieve interval to 10-15s when the values are low, so, when the block is the only thing keeping the card active; the longer interval becomes long enough to allow it to go to sleep. I made a script and use a custom block instead of this one.