[mythtv-users] How much horespower for frontend?

[Brian Wood]

On Sep 27, 2006, at 6:38 PM, James Pifer wrote:

>> I actually can't guarantee that playback of recordings is OK. I'm
>> currently trying to get my nvidia card re-installed back into my
>> combined FE/BE since it's where I really want it to run. That's the
>> reason for my current silence.... I will respond to the posed  
>> questions
>> soon I hope.
>>
>> I will try BoB though ( I assume that's in mythfrontend's setup
>> somewhere) ?
>>
>> Thanks,
>> James
>
> Well, not much good news to report. Still have the stuttering problems
> during playback. Here's what I've tried with the suggestions from
> everyone.
>
> 1) DMA is enabled AND set to 32 bit.
>
> 2) Enabled and Disabled "Enable realtime priority threads";no
> difference;not running as root
>
> 3) Lowered recording profiles to 640x480 from 720x480; no change
>
> 4) De-interlacing works best with "Kernel (less motion blur)";BoB was
> the worst for me
>
> 5) Comm Flagging is after recordings
>
> 6) CPU still has not gone above 40% and is usually low 30's
>
> 7) XvMC; How do I know if this is enabled? I'm really not sure about
> this.
>
> 8) ****Playback of recordings also has the same issue****
>
> 9) MediaMVP plays recordings perfectly (and honestly the picture  
> looks a
> little better, it's just missing the functionality I want from the
> frontend)
>
> 10) What about mythvideo? I get stutters there to. If I do finally  
> find
> a solution, can it/does it apply to MythVideo when it's set for the
> Internal player?
>
> 11) I'm still open to trying the Celeron machine to see if it does a
> better job. Based on what I've seen here what I have should work.
> UNLESS, the PCI video card is the problem. In that case my luck isn't
> changing as the Celeron machine does not have AGP or PCIe either... :-
> ( Plus, my BE had four hard drives and I don't think the Celeron  
> machine
> has enough 3.5 bays, so there would be lot's of work there too.


Well Bob is the most CPU-hungry de-int method, so it really is  
starting to sound like you might be starved for CPU, but I'd expect  
the playback to look best with no de-int if that was the case,  
ignoring interlace artifacts that is.

Unless, as you say, the PCI video is the real culprit, which may well  
be the case.

Have you read the reference at the end of the PCI latency WiKi entry?:

http://www-128.ibm.com/developerworks/library/l-hw2.html

This might help you get a little more out of your PCI buss to PCI  
Video card transfer rate, especially if you have a VIA chipset.

Tweaking the latencies did wonders for me, going from choppiness with  
just a playback and a record to being able to record 4 programs and  
playback a fifth with no problems. This was on a box with plenty of  
CPU though, disk throughput was the problem in that case, and it was  
making poor recordings which you are apparently not.

But in the end I replaced the VIA-chipset mobo with an nForce board.  
A lot of other squirrels went away at that point.