[mythtv-users] More XvMC data points

[Tom Lichti]

Tom Lichti wrote:
> Michael Tiller wrote:
>   
>> On 11/29/06, *Michael Tiller* <michael.tiller at gmail.com 
>> <mailto:michael.tiller at gmail.com>> wrote:
>>
>>     Just out of curiosity...when you say 65-75% CPU utilization, how is
>>     that partitioned between the X server and the frontend?  I ask
>>     because if I try and watch without XvMC on my system, the X server
>>     is something like 30% and the frontend is like 70%.  I'm wondering
>>     if this load on the X server can be avoided somehow.  Then I might
>>     have a chance to watch HD natively. 
>>
>>
>> I just wanted to circle back on this issue of X server utilization.  I 
>> installed a 2.4GHz P4 in my machine to replace the Celeron that was 
>> there.  Now when I attempt non-XvMC playback, I get something like 47% 
>> CPU for the frontend and 50% CPU for the X server.
>>
>>  From all the data I have seen, this seems like extremely high 
>> utilization of the X server.  I just have this sense that perhaps there 
>> is something poorly configured here that it requires so much CPU from 
>> the X server.  I know the XvMC gives you access to the motion 
>> compensation framework but am I perhaps missing something more 
>> fundamental like some acceleration that would allow direct writes to the 
>> framebuffer or something?!?
>>
>> --
>> Mike
>>     
>
> I was doing some playing around with my main frontend last night, and 
> playing an HD signal without XvMC enabled, I had (on a dual core P4) 99% 
> usage on core one (MFE) and 56% usage on core two (X). With XvMC 
> enabled, I had about 45% CPU combined, so XvMC *is* working for me on my 
> main system. This was with current SVN as of last night, and the latest 
> nVidia beta drivers (9742). I had no stuttering, no prebuffer pauses, or 
> any other issues, aside from what *seems* like frame skipping, but I see 
> nothing in the logs to indicate that, so I'm not sure if it's real or not.
>
> My backup frontend I have not tried yet.
>
> Tom
>   
Oh, some more info, I compiled with --disable-xvmcw to disable the 
wrapper. Unfortunately that seems to have introduced a LOT of 
instability, so I am going to try with the wrapper and see what happens.

Tom