[mythtv-users] Where to put the power..

[Rich West]

Rod Smith wrote:
> On Sunday 11 March 2007 01:28, Rich West wrote:
>   
>> Simple question (or so I think)..
>>
>> Is it better to have a dramatically better processor in your FEs or in
>> the BE?  And is the answer different if you are doing SD vs HD?
>>     
>
> As with so many things, the answer to both questions is "it depends." You 
> might want to think of it in terms of the CPU power required to do certain 
> processes. As a very rough guideline, using Intel CPU speeds:
>
> - SD encoding with hardware encoding card: ~50MHz
> - Digital TV (HD) capture: ~100MHz
> - SD playback with XvMC: ~700MHz
> - SD playback without XvMC: ~1000MHz
> - SD encoding with frame grabber card: ~1000MHz
> - SD transcoding in 1:1 time (1 hour to do 1 hour): ~1500MHz
> - HD playback with XvMC: ~2800MHz
> - HD playback without XvMC: ~3500MHz
> - HD transcoding in 1:1 time: ~4000MHz (??)
>
> Some of these estimates are VERY rough; they depend on factors such as the 
> recording options used, driver efficiency, etc. Still, I believe they're 
> reasonable first-pass numbers, with the possible exception of the HD 
> transcoding value (I've done very little of that). Also, transcoding can be 
> done on systems with any CPU speed; it just takes longer on slower CPUs.
>
> For your frontend, locate the stream type you'll be playing back and get a CPU 
> in that range or faster. (XvMC is a feature supported by some video cards and 
> drivers, such as nVidia-based cards using nVidia's binary drivers.) For the 
> backend, locate the type of encoding you'll be doing, multiply by the number 
> of tuners, and get a system with at least that much power. If you expect to 
> do any transcoding at all, add in enough CPU power to do that in a time you 
> consider acceptable, or take the larger of the recording and transcoding 
> values if you can live with transcoding only when recording isn't occurring. 
> For a combined frontend/backend system, add all these values together. Note 
> that transcoding is usually required for burning DVDs, unless you use a 
> hardware MPEG-2 encoding card and record in the right format and bitrate to 
> begin with.
>
> Depending on your choices, you might need more power in either the frontend or 
> the backend. For instance, if you want three tuners, you'll need more backend 
> power than frontend power if you use software-encoding (frame-grabber) 
> tuners, but more frontend power than backend power if you use 
> hardware-encoding cards and if you don't need extremely zippy transcoding 
> speeds.

Thanks for the responses!  Just to follow up, I did search the 
gossamer-threads, but the discussions I came across specifically covered 
the amount of horsepower (Ghz/Mhz) needed for the frontend.

For what it is worth, I have a backend with three PVR-150's on a AMD 754 
Athlon64 3000 with only 512MB of RAM (I have another 512MB stick I am 
considering throwing in there, but I just don't see the need just yet) 
and three frontends with AMD 754 Sempron64 3000's and 512MB RAM (each).  
All of the systems seem to be rather bored with SD (especially the BE), 
and, after seeing some recent posts about *not* using the Sempron 3000's 
with HD, I started to get curious..

-Rich