Open arcivanov opened 4 years ago
This is the example video that allows to reproduce the issue and the workaround 100% of the time.
@arcivanov we fixed some vp9 issues in master branch. If possible, you may give a try with master branch. As for 8K, we won't add support for 8K in this driver.
i like use the small monitor more easy this is old machine and old model of monitor television machine can slow because resolution
I can just add my support to everything of what @arcivanov said.
A build of chromium-freeworld (the one with VA-API patch) + latest release from here + go to any Youtube 4k video, right-click and choose "stats for nerds" = Dropped frames
The same, but with intel-media-driver instead = No/very few dropped frames
Something is not right with intel-vaapi-driver!
For years I thought that the reason I couldn't watch VP9 4K@60Hz on YouTube was first that my Skylake was underpowered. But then I upgraded to Coffeelake and Fedora 31 and kernel 5.3.8 and I still would lose 30% of all the frames and would experience constant stutter. So I thought it was chromium-vaapi/-freeworld that was the problem due to bugs and constant VAAPI module rewrites in the upstream, or slow rendering etc.
Today I tried iHD and all my issues went away - frames dropped are in single digits, not a single stutter, CPU consumption is much lower. So the issue is definitively an intel-vaapi driver. It's 100% reproducible and the only difference is whether chromium is launched with
LIBVA_DRIVER_NAME=iHD
.The question is why? Are there any profiling tools in libva/vaapi toolkit that would allow to see the decoding/rendering pipeline to determine the difference?
Below is some info about the system and screenshots of the chrome://gpu video capabilities. As you can see iHD reports VP9 decoding to up to 8192x8192.
i965:![image](https://user-images.githubusercontent.com/5355666/68081927-b950e200-fde3-11e9-8a74-554f7523983e.png)
iHD:![image](https://user-images.githubusercontent.com/5355666/68081922-a5a57b80-fde3-11e9-9506-206a00f1a392.png)