Mailing List Archive

HDTV on Myth -- not a "myth" CONCLUSION
First off, I want to thank everyone who shared
information and enlightened me on this topic. As
always, I have learned a tremendous amount from the
keep knowledge well of this list.

Here are some facts I have learned:

1. Xv is more tested and debugged than XvMC. It has
been stable for much longer and is preferred for its
flexibility (e.g. filters) and its integration with
the video architecture of Myth.

2. Xv has higher computational load than XvMC. A
marginal processor won't cut it. The hyperthreading
of the P4 is an advantage, but it isn't clear yet how
much. I know my Athlon 2500+ is not up to the task of
decoding 1080i material on Xv.

3. The video card makes a tangible difference. I
tried an nVidia 5200 card as a drop-in replacement for
my GF4MX card, and I was able to smoothly decode 1080i
on Xv, but only while there was no commercial flagging
going on in the background. One other difference in
the card was AGP 8x in the 5200 vs. 4x for the GE4MX.
Neither card chose to use fast writes, however.

4. Factors other than raw processor speed can affect
the system performance. For example, the PCI
backplane can get "swamped" even though the processor
is at less than 100%. This can happen due to HDTV
streams being recorded from one or more cards while
being played back from the disk. HDIV performace is a
system issue, not any individual component.

5. Consider separating FE and BE processes to
eliminate tasks from affecting one another. Take my
example of commercial flagging for instance. Move it
to a separate machine and it won't interfere with
playback on Xv.

6. Don't use alsa 1.0.6 drivers, they have been
implicated in video jerkiness. Do enable "extra audio
buffering." Do use gcc 3.4 instead of 3.5 for
slightly better performance (what about earlier
releases? I'm on gcc 3.3)

7. If stability is of highest importance, match the
hardware and software used by the developers of HDTV
on Myth. Use the Pentium 4, enable HT, and choose Xv
on an nVidia card. If you want to contribute and help
debug new platforms, choose something else. ;-)

I hope this adequately summarizes what I've learned in
the last couple of days. I'm limping along with my
Athlon 2500 and GeForce 5200 card right now, but
planning to replace the mobo with a more capable one
and dedicate this one to a BE server with lots and
lots of HD-3000 cards and stacks of cheap Fry's hard
disks (I wish).

Joe Barnhart





_______________________________
Do you Yahoo!?
Declare Yourself - Register online to vote today!
http://vote.yahoo.com
Re: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
Joe Barnhart wrote:
> First off, I want to thank everyone who shared
> information and enlightened me on this topic. As
> always, I have learned a tremendous amount from the
> keep knowledge well of this list.
>
> Here are some facts I have learned:

[snip]

Thanks, Joe. Not trying to incite any Wiki discussions here, but you
might want to see about getting this posted to www.mythtv.info .

-Doug
Re: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
On Tue, 12 Oct 2004 07:32:50 -0400, Doug Larrick <doug@ties.org> wrote:
> Joe Barnhart wrote:
> > First off, I want to thank everyone who shared
> > information and enlightened me on this topic. As
> > always, I have learned a tremendous amount from the
> > keep knowledge well of this list.
> >
> > Here are some facts I have learned:
>
> [snip]
>

Silly question,

Do I need to do anything special to enable XV instead of XVMC? I know
there's a settings.pro part, but is there a different package/library
I need for it?

Thanks
Re: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
A vanilla mythtv source should use Xv if there. Run 'xvinfo'
while running X to make sure it's going. If not, it's a "Module extmod"
or something in XF86Config.


*************************************************************************
* Cory Papenfuss *
* Electrical Engineering candidate Ph.D. graduate student *
* Virginia Polytechnic Institute and State University *
*************************************************************************


On Tue, 12 Oct 2004, Art Morales wrote:

> On Tue, 12 Oct 2004 07:32:50 -0400, Doug Larrick <doug@ties.org> wrote:
>> Joe Barnhart wrote:
>>> First off, I want to thank everyone who shared
>>> information and enlightened me on this topic. As
>>> always, I have learned a tremendous amount from the
>>> keep knowledge well of this list.
>>>
>>> Here are some facts I have learned:
>>
>> [snip]
>>
>
> Silly question,
>
> Do I need to do anything special to enable XV instead of XVMC? I know
> there's a settings.pro part, but is there a different package/library
> I need for it?
>
> Thanks
>
Re: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
On Tuesday 12 October 2004 00:29, Joe Barnhart wrote:
> First off, I want to thank everyone who shared
> information and enlightened me on this topic. As
> always, I have learned a tremendous amount from the
> keep knowledge well of this list.

Me too! :-)

> 2. Xv has higher computational load than XvMC. A
> marginal processor won't cut it. The hyperthreading
> of the P4 is an advantage, but it isn't clear yet how
> much. I know my Athlon 2500+ is not up to the task of
> decoding 1080i material on Xv.

D'oh. You have a 2500? I could have sworn you had a 2800. That makes a whole
lot more sense now. The 2500 runs a lower system bus speed (333 max, IIRC),
and you've likely only got 256k on-chip cache, versus my 3200, which is
running on a 400MHz system bus with 512k on-chip cache. I think the cache is
likely a big win for HDTV. I've got a system with a 2600/333/256 in it, maybe
I'll give that a shot at some point for comparison's sake... You must be
drilling the cpu pretty good in Xv mode...

> 3. The video card makes a tangible difference. I
> tried an nVidia 5200 card as a drop-in replacement for
> my GF4MX card, and I was able to smoothly decode 1080i
> on Xv, but only while there was no commercial flagging
> going on in the background. One other difference in
> the card was AGP 8x in the 5200 vs. 4x for the GE4MX.
> Neither card chose to use fast writes, however.

Good to know. I'm glad I decided to get a 5200 for the hell of it when I got
the new mobo for my box. :-)

> 6. Don't use alsa 1.0.6 drivers, they have been
> implicated in video jerkiness. Do enable "extra audio
> buffering." Do use gcc 3.4 instead of 3.5 for
> slightly better performance (what about earlier
> releases? I'm on gcc 3.3)

I believe that should be use 3.4 instead of 3.3 for slightly better
performance. I'm not even sure there's been a 3.5 release yet, 3.4 is pretty
bleeding edge. I'm going to try moving my box to Fedora Core 3 next weekend,
which includes 3.4, so I'll finally rebuild Myth w/3.4 at that point.
Definitely interested to see if I get any performance gain...

> 7. If stability is of highest importance, match the
> hardware and software used by the developers of HDTV
> on Myth. Use the Pentium 4, enable HT, and choose Xv
> on an nVidia card. If you want to contribute and help
> debug new platforms, choose something else. ;-)

One of these days, maybe I'll try swapping in a Radeon video card for
giggles... I have too many "one of these days" projects...

> I hope this adequately summarizes what I've learned in
> the last couple of days. I'm limping along with my
> Athlon 2500 and GeForce 5200 card right now, but
> planning to replace the mobo with a more capable one
> and dedicate this one to a BE server with lots and
> lots of HD-3000 cards and stacks of cheap Fry's hard
> disks (I wish).

:-p

One of these days, I'll also get around to finishing the HDTV page on my web
site. Pretty much all of this is good info for it (as well as the mythtv.info
wiki, as Doug mentioned).

--
Jarod C. Wilson, RHCE
jcw@wilsonet.com

Got a question? Read this first...
http://catb.org/~esr/faqs/smart-questions.html
MythTV, Fedora Core & ATrpms documentation:
http://wilsonet.com/mythtv/
MythTV Searchable Mailing List Archive
http://www.gossamer-threads.com/lists/mythtv/
Re: Athlon processors was: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
--- Jarod Wilson <jcw@wilsonet.com> wrote:

> I know my Athlon 2500+ is not up to the
> task of
> > decoding 1080i material on Xv.
>
> D'oh. You have a 2500? I could have sworn you had a
> 2800. That makes a whole
> lot more sense now. The 2500 runs a lower system bus
> speed (333 max, IIRC),
> and you've likely only got 256k on-chip cache,
> versus my 3200, which is
> running on a 400MHz system bus with 512k on-chip
> cache. I think the cache is
> likely a big win for HDTV. I've got a system with a
> 2600/333/256 in it, maybe
> I'll give that a shot at some point for comparison's
> sake... You must be
> drilling the cpu pretty good in Xv mode...

I tried a spare 2800 from another system for awhile.
It made little difference by itself. Both the 2500
and 2800 are "Barton" cores and run at 333FSB with
512K cache.

I also have a "spare" AMD64-3200 and an nForce3 mobo I
could throw into the system. Hmmm... I wonder if I'd
have to reinstall Linux or if it's close enough to
just put it in and boot... I'm using KnoppMyth and I
have not recompiled anything for processor type (yet),
I'm just using the stock i686 debs for Linux and Myth
(except I did recompile mythfrontend).

Maybe I'll back up and just give it a go. I'm not
feeling enough pain yet... :-)


__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
Re: Re: Athlon processors was: HDTV on Myth -- not a "myth" CONCLUSION [ In reply to ]
If you do recompile try the CFLAGS "-Os -march=athlon-xp" on both your
kernel and your mythtv installation. You'll find that in
arch/i386/Makefile for the kernel and you just need to add the
optimizations to the settings.pro file for mythtv.

That's what I use on all my linux and myth boxes and they seem to run
just fine, though I'm not hardcore like you guys and have no HDTV.
There's anecdotal evidence on the linux-kernel mailing list that
optimizing for cache usage improves performance system wide.

GCC Manual:
http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#Optimize-Options

-Os
Optimize for size. -Os enables all -O2 optimizations that do not
typically increase code size. It also performs further optimizations
designed to reduce code size.

-Os disables the following optimization flags:

-falign-functions -falign-jumps -falign-loops
-falign-labels -freorder-blocks
-freorder-blocks-and-partition -fprefetch-loop-arrays


Joe Barnhart wrote:

> I'm using KnoppMyth and I
> have not recompiled anything for processor type (yet),
> I'm just using the stock i686 debs for Linux and Myth
> (except I did recompile mythfrontend).
>
> Maybe I'll back up and just give it a go. I'm not
> feeling enough pain yet... :-)
>
>