06:49 notthatclippy[d]: ristovski[d]: That's not quite what the pstate is. You can have every single piece of the GPU active at P0 just like at P8, it's just that the base and memory clocks will be lower.
06:51 notthatclippy[d]: For this particular issue my bet would be that there is something that is periodically waking the GSP now that wasn't doing it previously. If you're on a recent proprietary driver, you can send me the `nvidia-bug-report.sh` output and I'll see if there's an obvious culprit.
06:57 ristovski[d]: notthatclippy[d]: Ah, I had assumed that P8 would be special in the sense it would indicate a truly "idle" state, but I guess that assumption is wrong
06:58 notthatclippy[d]: There's a separate truly idle state (well, multiple actually), called GC6
06:58 notthatclippy[d]: In that state most of the GPU is off, including the GSP, and there's only a few power islands around.
06:59 ristovski[d]: I see, safe to assume that the state of it isn't exposed to userspace?
06:59 notthatclippy[d]: I think it is, somehow, somewhere
06:59 notthatclippy[d]: Gotta run now, I'll check later.
06:59 notthatclippy[d]: btw, this is a related issue FYI: https://github.com/NVIDIA/open-gpu-kernel-modules/issues/693