[mythtv-users] Oddly VDPAU makes things WORSE

Andrew Stevens wackston at googlemail.com
Thu Dec 10 22:45:20 UTC 2009

Update: looks like was compositing was underlying issue.

Disabling compositing made everything run smooth enough that shortish
test-runs are as good as I remember Myth ever being*. Good enough at
least to try for a few nights 'production' TV viewing.

Presumably the latency of the driver (or even X server??) responding to
the Damage event of the updated video plane window and blitting it into
place was enough to occasionally miss VBlank. 

Depatching VDPAU renderings as well presumably made it that much more
likely that stuff would be delayed enough to produce a visible stutter.

> Note, also, that if your processor is one where the memory bandwidth 
> drops when processor frequency scaling drops the CPU frequency, it can 
> make VDPAU not work.  This would include the AMD line of processors...

Messing around with the CPU frequency governors never seemed to make a
difference.  Mind you, my 9500GT is a discrete PCIe card - I can
certainly imagine slowing the integrated memory controller clock would
do bad things to the performance of a GPU using shared main memory.


* I used to develop DTV / TFT display / video processing ICs for a
living so I'm kind of sensitized to even very occasional frame-stutter.
There's a limit to what you can realistically expect from a stock linux
kernel and normal thread priorities.  

More information about the mythtv-users mailing list