I'm trying to get HD playback to work using a 3GHz P4 and a
band-new PCI FX5200 on an R5F27 system using "latest" nvidia
drivers (1.0.9755-1). No matter what I do, I can't get rid of
the "NVP: prebuffering" messages and corresponding audio
stutter.

Without chroma-key OSD, it's quiet hopeless: playback starts
smoothly with about 50-60% CPU usage. After 5-10 seconds of
playback CPU usage goes to 100% and both video and audio
playback stutter badly. Anything that causes the OSD to come
up (B/W) will also cause it to go into 100% CPU mode and
playback virtually stops. After the OSD goes away, it gets
slightly better -- playback resumes but it stutters very badly
and CPU usage stays at 100%. I tried all combinations of
events, sync, and texture settings I could think of in both
nvidia-settings and xorg.conf.

With chroma-key OSD enabled, it's better: playback runs
smoothly at 50-60% CPU usage with occasional bursts of "NVP:
prebuffering messages" and some barely noticeable audio
stutter. When the OSD comes up (in color), playback continues
with 100% CPU usage and mild stuttering. After the OSD goes
away, the high cpu-usage and stuttering continues for a few
seconds and then goes back to 50-60% with bursts of "NVP:
prebuffering"

I also trying enabling/disabling video-timebase and extra audio
buffering. Neither made any difference. Enabling/disabling
deinterlacing didn't make any difference. I'm using the TV-out
at 640x480 if that makes any difference.

Playback of SD with xvmc works perfectly with CPU usage of
about 10%.

After spending a couple hours reading forum and newsgroup
threads I've come to two conclusions:

1) MythTV's xvmc playback support went into the toilet with
0.20. I found a lot of people reporting that xvmc worked
fine for them with 0.16/0.18, but upgrading to 0.20 broke
xvmc completely. Some reported that in more resent SVN
snapshots xvmc was working again.

2) Older nvidia drivers seem to have fewer problems.

Downgrading MythTV isn't an option (I need HDHR support).

Downgrading the nvidia driver is an option.

Building MythTV from SVN is an option.

Can anybody recommend a combination of nvidia driver version,
settings, and MythTV version that works for xvmc playback with
an FX5200?

Would upgrading from 512MB to 1GB help?

_________________
Grant

Try upgrading the RAM. I just don't think 512mb is enough for HD. But I think you have other problems. I started with r5e50 (.20) with a slower cpu (sempron 2800+) and never had the problems you're having while XvMC was enabled. CPU usage of 10% is high with XvMC running.

Also look at your graphics card. I have a 6200 and I think I'm running 9755 drivers. How much RAM does your card have? Mine has 256mb, and I think I've seen some that have no RAM at all, that "steal" RAM from the system memory. I've read some people think the 5200 is a better unit, so this might be an issue also.

The "magic bullet" for me was to add to my "Screen" section in XF86Config-4 (I'm still running R5D1, don't know if this will help you with Xorg).

Similar specs to yours (P4 3.0, 512Mb RAM) but using a 6200LE. nVidia driver version 9755, not using XvMC.

The Screen section? I tried adding it to my Device section,
and when set to true, it made things worse: About 75% of the
time the video overlay just comes up black when you hit "play".
When it does show video, the stuttering is worse and CPU usage
is higher.

Everything I've read leads to believe that my setup ought to be
workable. It's a PCI board rather than an AGP board, so that's
one difference, but I've read of others with PCI video boards
that seem to have HD playback working fine.

I suppose I could buy a 6200 board and try that, but I was under
the impression that a 6200 wouldn't be any better for xvmc and would run hotter.

_________________
Grant

I'll give that a try. I guess I can always put the RAM in a
differnt machine if it doesn't help my Myth box.

As do I. I suspect some of it is due to using PCI instead of
AGP (the board doesn't have an AGP slot), but others report
similar PCI boards are working fine.

128MB on a 64 bit bus.

If I knew a 6200 would work, I'd order one, but nobody seems to
think that a 6200 helps more than a 5200.

_________________
Grant

just another data point for you:

AGP FX5200 128 Mb using TVout
2 Ghz P4 cpu
512 Mb ram
xvmc enabled
R5F27 clean install
combined backend/frontend
Don't know what chroma-key OSD is or where you turn it on or off. so mine is what ever the default install is.

This plays back OTA HD recordings fine (or watching live HD). Stutters slightly while OSD is up, but tolerable and is fine as soon as OSD is gone.

This system is way slower that yours with the same amount of ram. Kinda looks like the difference may be AGP vs PCI. Unless your hard drive is not DMA enabled or slow for some other reason, like being very old and/or small.

If the little pop-up displays while watching TV (like when you
skip forwar or backward) are B/W instead of color, then you're
not using Chroma-key OSD. To enable the chroma-key OSD feature
you have to poke a value into the mysql database by hand:

http://www.mythtv.org/wiki/index.php/Xv ... 4_.26_5.29

For whatever reason, doing so produced a huge improvement in
stability for me.


It must be. I'll have to try to find the threads where people
said PCI was working for them. It could be they had a video
board with more horsepower or faster/more CPUs.

DMA is enabled on the drive, and it's a 200GB Samsung Spinpoint
that's less than 2 years old.

_________________
Grant

Some figures I found via Google:

PCI 2.2 = 133mb/sec
AGP 1.0 = 266mb/sec
AGP 2x = 533 mb/sec
AGP 4x = 1.06gb/sec
AGP 8x = 2.1 gb/sec

The 6200LE I referenced above is an AGP (I missed that you mentioned your 5200 was PCI) in a 4/8X AGP slot. I'm not positive but I believe the card is 4x.

I had stuttering when I first attempted HD using a test system with a 2x AGP slot, Celeron 2.0Ghz, 512Mb RAM and FX5500. The 5500 is now in another non-Myth system as I purchased the 6200LE thinking that was going to solve my problems - NOT! So yes, the other components are critical to smooth playback as well. I, like jzigmyth, experience stuttering only while the OSD is displayed (color OSD).

If people are viewing HD via a PCI card I suspect there's a lot of tricks to be played...

Thanks for the chroma-key tip! My OSD is B+W with xvmc enabled. I'll have to try the hack. Your hard drive sounds like it is up to snuff.

That looks right. Vanilla PCI is 33MHz X 4bytes. I'm pretty
sure the my Shuttle box's slot is the 33MHz variety, but
neither the manual nor the spec sheet actually says. It does
say that the on-board ATI 9100 is on an 8X AGP bus. Too bad
there's no support in Linux for the 9100's MPEG decoding
features.

I'll have to do some more searching on PCI and HD playback. I'm
starting to get the feeling that my little Shuttle Zen just
isn't going to be able to do HD playback. :/ Too bad it didn't
have an on-board NVidia video chip instead of that useless ATI
one.

Maybe it's time to order a socket AM2 microATX board with an
on-bord NVidia 6150.

_________________
Grant

That's what I ended up doing for 2 of my frontends - Asus M2NPV-VM. Love 'em, but they're not microATX

Yes they are. Unless this page is wrong:
http://www.asus.com/products4.aspx?l1=3 ... odelmenu=1

_________________
Grant

According to lshw the PCI clock for that slot is 66MHz, but
that's still a rather sluggish 266MB/s. :/

_________________
Grant

Hmmm, so they are at 9.6 x 9.6. I couldn't fit one in a (supposed) microATX chassis so I assumed (yeah, I know) that they weren't truly microATX. Must have been the case, apologies for the mis-information.

Hmmm,
so they are at 9.6 x 9.6. I couldn't fit one in a (supposed)
microATX chassis so I assumed (yeah, I know) that they weren't
truly microATX. Must have been the case, apologies for the
mis-information.[/quote]
From what I can gather, there are a lot of incompatabilies
between various microATX boards and microATX cases -- usually
due to locations of CPU and RAM interfereing with other stuff
in the case. I've also seen supposedly standard ATX boards
that just plain weren't the correct dimensions.

_________________
Grant