[plugh]

I am currently using
avisynth 2.56,
with
dgmpgdec148
decomb522_25_dll_20050904
colormatrix_25_dll-src_20050223
degrainmedian082 (as needed)
and a patched version of
XviD-1.1.0-30122005
with
virtualdubmod 1.5.10.2

My xvid patch is to further tweak the built-in "HDTV" profile for xbmc (turned off 'packed bitstream' and am still fine-tuning the "video buffer verifier" values)

Haven't tried MeGUI - doesn't it use xvid_encraw? I believe I read that encraw provides command line level access to the vbv parameters...

[teldin]

Cound you post one of your scripts plugh?

[plugh]

There really isn't a 'default' avisynth script - it's highly dependant on source characteristics. I may post my tweaked xvidvfw.dll sometime, after I'm satisfied with my tuning of the profile parameters (lots and lots of encode / playback tests - I'm spending WAY to much time on this).

[teldin]

Believe me, me and the whole HD Xbox community truly appreciate your hard work!

So to get me started, if my source is 1080i TS my .avs would look like:

Loadplugin("D:\AVIsynth 2.5\plugins\mpeg2dec3.dll")
Loadplugin("D:\AVIsynth 2.5\plugins\decomb521.dll")
Loadplugin("D:\AVIsynth 2.5\plugins\simpleresize.dll")
Loadplugin("D:\AVIsynth 2.5\plugins\undot.dll")
mpeg2source("c:\movie.ts")
Telecide(order=1,guide=1,post=2,vthresh=30)
Decimate()
crop(8,6,-8,-10)
UnDot()
simpleresize(960,720)

Does that look right?
Do I need to mess with a d2v file?

[plugh]

Get the dgmpgdec package - http://neuron2.net/dgmpgdec/dgmpgdec.html
Use it (dgindex.exe and dgdecode.dll) instead of mpeg2dec3.dll
and yes, it'll generate / use a d2v file.

telecide/decimate - whether to use them, and with what args is purely a function of your source stream. DGindex will give you some clues (if it is telecined film, you probably want "cycle=5" in the decimate call).

[plugh]

I've posted one of my HD Xvid encodes on demonoid. Search for "4xbmc".

This is a 1.5hr long 4GB test clip for which I am seeking playback reports on a variety of xbox/xbmc hardware/software configs so I can verify and/or further refine the xvid 'xbmc-hd' profile settings I have developed.

Side note to xbmc devo's - You might find this an interesting test clip for the dvdplayer...

[ultrabrutal]

Cool plugh! Good work. I'm all over it and will test it as soon as it's down

[ultrabrutal]

Hello plugh

I now got the movie downloaded and tested.

My Xbox is a version 1.4. Running T3CH build of XBMC from 2006-10-26.

I disabled FTP, Web, Autodetection and KAI. Then I watched the movie from my G-drive. However I forgot to increase cache until 30 min into the movie at this point I had 432 dropped frames. I also had soften filter on which I disabled.

When the movie was over I had 1707 dropped frames and close to 8000 early/late frames.

It appeared to have drops in audio often. I also noticed this at the credits where there were no frame drops but the audio dropped mean anyways. Weird

I hope you can use this info for something. I have a version 1.6 and 1.3 Xbox too. However they both have stock HD currently so I'm not sure there is enough room to put the movie there. If noone else gets back, I will see if I can let it play there.

Question: was the broadcast only in AC3 2.0 or did you choose this to get higher possible bitrate for video part?

[plugh]

Wow! It plays *perfectly* on my system. Now the hunt begins - what are the factors that gave the differant results...

I'm running a 4 month old xbmc build, on an older xbox, from a smaller disk (only an F drive). Guess it's time for me to try a more current xbmc build...

Yes, I would be interested in tests on your other boxes; the file should fit on an unclutterred E drive. 1.3/1.4/1.6 have differant video encoders, as I recall. Are your boxes soft or hard modded?

Didn't even think about soften / flicker filters - yes I have both off. Are you using the xbmc 'video alignment'? (I'm straight 1:1 mapping).

re: audio - 2.0 is what was in the transport stream. There are a few glitches in it (half dozen packets with crc errors, as I recall). Were you using software or external decoding?

I expected some variation, but not of that magnitude. Your result implies a huge slowdown somewhere. I'm truly astonished. Hope some of the other dozen plus people who grabbed a copy post their results...

[jeff_8]

Well, I kinda followed plugh's suggestions, but basically tried a few different things... Here's one that I'm doing right now, though there are other settings you could set as well:

I have a mkv file as source, so I ran MKVExtractGUI to get the audio and video separated, though I think I just need the audio (AC3) from this. Then opened MeGUI and clicked Tools - AVI Synth script creator. I chose my video file, and for the resize box typed 1120 and hit 'suggest resolution' to fill in the other number, then hit ok. Then it brings you back to the MeGUI main screen where I changed the video codec to XVid and hit 'Config'. Clicked 'turbo', picked 'Auotmated 2pass' and changed the bitrate to 1400 (See how that works, not sure about that one). Changed VHQ mode to Wide Search, hit ok, hit enqueue, then on the queue tab, hit start. When that's done, I just have to use the AVI Muxer tool to put the video and audio back together and it should be done! There are plenty of other options in the XVid codec config to play with, but these options should work, we'll see

[plugh]

That's a pretty low bitrate for HD. The one I posted had a target average bitrate of 5600kbps. Another one I'm working on looks like it'll be around 6300kbps.

I'm still scratching my head on 'brutals result. I just played the file again, with software audio decoding AND via SMB rather than from local hard disk. Had 1600 early / 1500 late frames (out of 137500) but still NO frame drops. Tried again, this time also enabled the ftp server and started copying files while playing via smb - finally got it to drop frames!

Definately got to test that 10-26 build tomorrow...

[ultrabrutal]

Maybe some devs could comment on which options is affecting playback in a negative way.
I think what they could implement was, when playback of 720p clip, XBMC will store settings in an alternative place, set the optimal/minimal settings for 720p, start the movie. When movie (and maybe also when paused) is stopped settings are restored. If XBMC crashes and restarts, it should see that alternative settings was stored, and thereby know that it crashed while playing back 720p, then it should restore the settings.

We have been talking about:
Servers
Video filters
Video decoders
Audio decoders

Maybe some of these does not affect performance at all. Possibly there are more options to consider?

Today I will test my two other boxes and get back with results

[ultrabrutal]

Test result:

Crystal Xbox - version 1.6
Softmodded with nDure1.1.1
Stock harddrive
XBMC T3CH 2006-10-16

Config:
Servers
FTP: enabled
Web: disabled
Auto detech: disabled
KAI: enabled

Video
720p 16:9
High Quality Pixel Shader
Framerate conversion: disabled
Soften: disabled
Flicker: disabled (0)
Calibration: (52,23) (42,23)
Cache: 2048

Audio
AC3 passthrough

Auto temp control: on at 55 degrees
Harddrive logging: enabled

SMB share: 67 drops (1817/1658 early/late)

67 drops is not bad I guess, but I think the goal is 0 plus having as few early/late as possible.. Also playback via network is a must I think. I have 3 boxes and looking around to buy/build a NAS dedicated to these... On to the next test

[plugh]

SMB share: 67 drops (1817/1658 early/late)

That's much more in line with what I'd expect.
Would be interesting to rerun it with
FTP off, Kai off, Temp = fixed
(I assume the RSS and uPNP is off as well?)

Is Ndure the one with the virtual C drive? I wonder how much of a hit that incurs. I'm using an older nkpatcher based softmod.

Regarding your 1.4 box test - I think there is something really off about that result. A couple thousand drops from hard disk on the 1.4 box vs 67 drops via SMB on the 1.6 box? We're missing something here...

----------------------------------------------------------

I ran a comparison between the 6-25 build I've been using and the 10-26 build. I enable mplayer benchmark output and xbmc logging, and played two clips under each build. First was a 700MB 42+ minute "HRHD" (960x528@24fps) avi, second was a 550MB 10+ minute HD test clip (~6300kbps 1280x544@24fps) I had previously encoded. Both were played from the F drive. All servers / services were disabled. Neither incurred any frame drops.

I'll post the full data if anyone wants, but here are the relevant results:

Code:

06-25 HRHD VC: 38.5271% VO:  2.7471% A:  4.1859% Sys: 54.5340%
10-26 HRHD VC: 40.0535% VO:  2.6925% A:  4.2203% Sys: 53.3034%
delta      VC: +1.53%   VO: -0.05%   A: +0.03%   Sys: -1.50%

06-26 myHD VC: 61.6298% VO:  2.9684% A:  3.7397% Sys: 31.6621%
10-26 myHD VC: 62.7203% VO:  2.7024% A:  3.7715% Sys: 30.8058%
delta      VC: +1.09%   VO: -0.27%   A: +0.03%   Sys: -0.86%

Conclusions -
Video decoding has gotten a bit slower
Video output has gotten slightly faster
AC3 5.1 software audio decoding just a tad slower

It is interesting that with my higher res clip the VC slowdown isn't as great - perhaps due to my (successful?) efforts to reduce decoding complexity in my encodes. It is also interesting that the slight VO speedup becomes more apparent at higher res (which makes sense).

Bottom line is that 10-26 is slightly slower overall than 06-25, but not significant enough to be the cause of brutal's first test result.

[jeff_8]

Re-encoded 1120x624 at 5000kbps and it works and looks great! (after copying to the hard drive, seemed a bit skippy over the network) By the way, I mentioned previously that I used the muxer tool in MeGUI to mux the video and audio when it was done, but that didn't seem to work for some reason, so I used AVIMux_GUI which worked great.