True.
Understood.
I'll try to get some soon.
I believe the T400 is a non-consumer card and doesn't have the p2 lock on it for CUDA workloads. The weird thing is that this happens even with only plex which I didn't think was considered a CUDA workload. (Although I don't know enough about it.)
According to "nvidia-smi -q -d SUPPORTED_CLOCKS | more" my max clocks on my 3050 should be 7001 and 2130. Under power state p2, I am only getting 6800 and 1807, so it is definitely being slowed. If I try to change it using "nvidia-smi -ac 7001,2130", it tells me that it isn't supported.
The last 8 episodes of a show I transcoded (the originals are all in the same h.264,. 1080p format)
1060TI (on remote unRAID server) average FPS:
81,83,87,82
1660 (on local unRAID server)Super:
58.60.57.58
I have enabled nvidia-persistenced.
It is definitely slower. I reliably get about 20-50% more fps on the 1060TI on similar files (running 2 workers on each machine).
When I use it in Plex or Tdarr docker containers, it never gets above P2 power state (as reported by GPU statistics). When it is not being used, it drops to P8 (which is normal).