2024-04-25, 10:04 AM
(This post was last modified: 2024-04-25, 10:07 AM by nurunet. Edited 1 time in total.)
So, I did some more experimenting. I think the Athlon 200GE was not a good test case because it seems to do all encoding in software, which works on the 4650G too.
I tried it on Windows. There, I had to use AMF. The results:
I also did some more testing on the 4650G. Often, it plays okay for some time before it starts to show artifacts, so I set a more aggressive fan curve, but that did not solve the issue. Even if the integrated GPU only reported ~35°C afterwards.
There seems to be a correlation between the number of new or concurrent transcodes and appearance of artifacts. As far as I can tell, skipping in a video leads to a new transcode starting at the new position, and doing that a few times increases the likelihood of artifacts.
Playing back more than one concurrent stream also leads to artifacts (and stronger ones) it seems.
So either there is a defect in my Ryzen Pro 4650G's video encoder unit, but I have no idea how to test for that, or an incompatibility between it and one component of the Jellyfin transcode chain - but I would have thought that the 4650U and 5600G should be sufficiently closely related to be affected as well.
I have run a Furmark stress test on the 4650G (w/ scan for artifacts, whatever that does), but I did not find any unusual behavior.
Any ideas, what i could test next?
P.S.: A typical case of a small artifact (the bird's wings, ca. at 03:00 seconds): https://dai.ly/x8xfjvo. And the same clip being played (and transcoded) on the 4650U w/o issues: https://dai.ly/x8xfjvm
I tried it on Windows. There, I had to use AMF. The results:
- On my Server's Ryzen 4650G, it was unwatchably bad
- On my notebook's Ryzen 4650U, it was fine (booted it off the same USB SSD as the server even)
- On my PC's Ryzen 5600G, it was fine
I also did some more testing on the 4650G. Often, it plays okay for some time before it starts to show artifacts, so I set a more aggressive fan curve, but that did not solve the issue. Even if the integrated GPU only reported ~35°C afterwards.
There seems to be a correlation between the number of new or concurrent transcodes and appearance of artifacts. As far as I can tell, skipping in a video leads to a new transcode starting at the new position, and doing that a few times increases the likelihood of artifacts.
Playing back more than one concurrent stream also leads to artifacts (and stronger ones) it seems.
So either there is a defect in my Ryzen Pro 4650G's video encoder unit, but I have no idea how to test for that, or an incompatibility between it and one component of the Jellyfin transcode chain - but I would have thought that the 4650U and 5600G should be sufficiently closely related to be affected as well.
I have run a Furmark stress test on the 4650G (w/ scan for artifacts, whatever that does), but I did not find any unusual behavior.
Any ideas, what i could test next?
P.S.: A typical case of a small artifact (the bird's wings, ca. at 03:00 seconds): https://dai.ly/x8xfjvo. And the same clip being played (and transcoded) on the 4650U w/o issues: https://dai.ly/x8xfjvm