Posts: 952
Joined: Feb 2005
Reputation:
0
plugh,
Yes ndure uses virtual C and nkpatcher .67.
On my 1.4 chipmodded box, I rerun part of the test with FTP off, Kai off, Temp = fixed (custom fan installed), rss off, upnp off. Everything I can turn off is off..
Just within the dreamworks logo I had 5 drops. when the timewarner logo went away I had 30. when the scene with the staircase begins i had 86 drops. Then I stopped playback... This was run via SMB
Can it be that a modification with Xecuter3 slows down compared to softmod? Can it be that a 1.4 box is slower than 1.3 and 1.6 possibly due to different video chip)?
I wish more people with different hardware versions would start benchmarking too
Posts: 294
Joined: Feb 2006
Reputation:
2
plugh
Senior Member
Posts: 294
Yeah, I'm kind of disappointed at the lack of playback reports...
Are you using a differant dashboard on the 1.4 box? Perhaps with some background function enabled? I'm using evoxdash - I'll have to check if TSR and/or IGR are enabled.
Does your hardmod allow you to switch to booting with a retail bios? If so, might be interesting to boot retail and apply an nkpatcher softmod.
Posts: 468
Joined: Oct 2003
Reputation:
0
Downloading the file now.. I'll give you info later.
Posts: 952
Joined: Feb 2005
Reputation:
0
plugh, XBMC is my default dashboard which can't be the problem. XBMC is default dash on all 3 boxes I have. IGR is enabled on all 3. The one with the problems is the only one which is hardmodded... I guess I could put back the original harddrive, disable the xecuter3, and install softmod on org. harddrive and see if it makes a difference... Will report back later.
When I downloaded the file, I saw alot of people doing the same. Don't know if they found the movie via a search function or via this forum. If the later, I am too, disappointed
Posts: 294
Joined: Feb 2006
Reputation:
2
plugh
Senior Member
Posts: 294
You are changing two (bios and hard drive) or three (xbmc version?) variables if you go that route. If you backup C, switch to retail, and apply softmod, then the only change is the bios/kernel.
Yeah, they may have simply found the file on demonoid but I was sort of hoping for reports to be posted in the torrent comments section or via PM, as I explicitly stated it was an experiment and requested feedback both in the posting and the nfo file.
Posts: 952
Joined: Feb 2005
Reputation:
0
Blasted... The original harddrive is unlocked and I cannot get into Config Live for some reason to get it locked again. I do not have eeprom backup either anymore (lost the harddrive with the backup a while back)... So I hope someone else has a hardmodded 1.4 box to test with.
On a side note I noticed that the video quality is better on the 1.4 box than the others. Shaper and more steady. One of the others was kind twiching.
Posts: 294
Joined: Feb 2006
Reputation:
2
plugh
Senior Member
Posts: 294
Cool - I look forward to the reports.
Unless playback is obviously screwing up, refrain from bringing up the 'info' window until the end credits. Based upon some measurements I did today, it appears that window 'costs' about 2% of the cpu while being displayed.
And FWIW, software audio decoding of ac3 5.1 'costs' another 3.5% to 4%. What you are downloading only has ac3 2.0, which was only 1+% as I recall (I don't have those numbers handy). External audio decoding is the way to go if you have the capability...
Some other interesting results for a particular test clip I'm using:
Compared to playing the plain avi file from the local hard disk,
playing it from within a rar file on the local hard disk costs ~3.5% cpu
playing it as a plain avi (not rar'd) via 100mbps SMB costs ~5.5% cpu
(there are also latency issues to deal with in this case).
Posts: 952
Joined: Feb 2005
Reputation:
0
I had problems with DVD menus and subtitles. I went to find out the problem. In the process I had to reinstall XBMC with all settings. Then the menus and subtitles works again. No idea what caused it.
Then I for fun tried your movie again and to my surprise it played fine (at least for 15 min then I stopped it). So it seems there is nothing wrong with my hardmodded xbox version 1.4 afterall. Actually I think it produces the best video quality for the 3 I have...
I will run the full test later on. Now I'm too tired
Posts: 294
Joined: Feb 2006
Reputation:
2
plugh
Senior Member
Posts: 294
My 'design center' has been playback from local hard disk with external audio decoding. When I was testing that particular encode, I did it with software audio decoding in order to add a fairly constant amount of additional processing as a 'safety margin'. That is, if it played it with software decoding with minimal impact, playing it with external just gave a bit more 'elbow room'. And I also did my testing with the info window displayed, which I have since discovered added its own 'not insignificant' overhead (and hence more safety margin) to the mix.
As it turned out, I could still play it via smb, even with software decoding, still without frame drops - though you could tell things were more strained as reflected in the early/late frame counts.
(Note however, that I have never actually observed any particular visible impact from early/late counts - probably due to the 24fps source vs 60fps display differential. i.e. some frames are already displayed twice and some three times, so what mplayer considers early/late has little visibility, I guess.)
So to answer your question - I'm interested in a variety of playback scenarios. And as you can see, you can make trade-offs - for example, playback via smb with info window off and external audio decoding is probably close to the same load as playback from local hard disk with software audio and info window displayed (ignoring smb latency / networking hiccups).
However, when I start tackling the higher macroblock/second encodes (2.20:1, 2.00:1, 1.85:1 frames) the margins are likely to get very slim. At this point I've got a some encoding parameters, including a range of vbv values, that seem to be effective at avoiding frame drops with little impact on quality. However it is also becoming very apparent to me that bitrate based controls do not correlate well with decoder cpu load. That is, 10K bits in one frame can incur a significantly differant cpu hit than 10K bits in another. So controls that constrain 'bits/second' only loosely constrain 'instructions/second'.
Posts: 468
Joined: Oct 2003
Reputation:
0
I just played it over 100 Mbit SMB against WinXP.
The result was around 1700 early and 1600 late.
I have the webserver, auto detection Disabled.
Video Output 720p 16:9. High Quality Pixel Shader.
Soften Off. Flicker Filter Off.
Auido Output Digital.
I did see maybe 2-3 quick glitches in the movie but nothing big. (Some black squares) On the whole it played very well.
Anything you would like me to try ?
- ASRock ION 330 OpenELEC XBMC Frodo.
- 47" LG HDTV1080p, AC3/DTS Receiver.
- 96" Epson LCD 1080p projector
- 2x Raspbery Pi with XBMC
in a few days...
