Universal binaries status and benchmarks

[xavier86]

Any luck on the ac3 issue major?

[major]

During the last few days I have been quite busy. I'm resuming friday on this and will hopefully find something during the week-end.

[jmazzi]

How did the weekend go? Anything new to report?

[major]

Still not solved but with the help of a giant box of Coke I have just secured, the issue should not resist much more time.

[jmazzi]

Hows that box of Coke coming?

[major]

I just finished the 10th bottle.

[jmazzi]

Is this what your office floor looks like http://k41.pbase.com/u/slug/large/240.33.jpg ?

[major]

I've made some progress, by isolating a case which doesn't show the ac3 problem, this should lead to a solution.

[jasenko]

I've made some progress, by isolating a case which doesn't show the ac3 problem, this should lead to a solution.

Thanks major!
I've been running ffmpeg with VLC patches inside the PPC ffmpegX bundle for awhile now and I am mighty impressed. Can't wait for a full blown UB/Intel support.
Cheers mate.

[jmazzi]

I've made some progress, by isolating a case which doesn't show the ac3 problem, this should lead to a solution.

Good work!

[major]

The mplayer/mencoder ac3 problem is solved. All encoding components are now properly compiled Universal with MMX/SSE optimizations and fully static. I now proceed to cosmetics and packaging for forthcoming 0.0.9w release.

[jmazzi]

The mplayer/mencoder ac3 problem is solved. All encoding components are now properly compiled Universal with MMX/SSE optimizations and fully static. I now proceed to cosmetics and packaging for forthcoming 0.0.9w release.

Great! How long do you anticipate that taking? I'm defiantly gonna be registering it.

On a side note, do you think you might add a RSS news feed for your site?

[major]

Between 6 and 146 hours, depending on unpredictable factors. I'll make a new site for version 1.0.0, which will include a RSS feed.

[jmazzi]

Between 6 and 146 hours, depending on unpredictable factors. I'll make a new site for version 1.0.0, which will include a RSS feed.

Alright, sounds good. Have fun

[erk]

Impressive numbers!

An interesting thing I have observerd recently with H.264 encoding, I have a Mac Mini Core Duo and an IBM Thinkpad R52 1.73Ghz that I use as an OS lab rat. On the Mac Mini I typically get about 17-19fps using Handbrake 7.1 x264 baseline profile encoding a movie 1 pass for a 700MB target size.

When I tried the same test DVD, on the Thinkpad running Acidrip with the x264 mencoder codec under FreeBSD 6 and Ubuntu linux 6.06 I got a lousy 4-5 fps. However yesterday I installed the OpenSUSE 10.1-RC1 linux distribution and tried the same test and got 19-20fps constantly. So how can there be a 400-500% speed difference on the same hardware, there must be some processor feature the other distros missed? Does that mean there is more potential in the Mac Mini than Handbrake is achieving? Like I would be possible to see 30fps out of ffmpegX universal?

[major]

Most differences will depend on how the binaries are compiled and optimized, as well as on the source video being encoded (format, size, bitrate, complexity). As said above, on an Intel iMac I observed about 40fps on a test clip for the ffmpegX x264 preset (DVD to H264 at 640x272).

[erk]

Most differences will depend on how the binaries are compiled and optimized, as well as on the source video being encoded (format, size, bitrate, complexity). As said above, on an Intel iMac I observed about 40fps on a test clip for the ffmpegX x264 preset (DVD to H264 at 640x272).

Quite right, sorry I missed that x264 test you did I only noticed the ones in bold at the top of the thread. 40fps is very good, the Mac Mini is slower but it will be in the right ball park. As you say the type of source video can have a huge effect on the results, I try to use the same DVD for my tests. I did notice on the linux version that some of the support libraries were tagged i586 and others were i686 in the SUSE linux, I presume they all used SSE in the x264 codec like you are doing.

For a comparison I am just installing the OpenSUSE 10.1 RC1 on the Mac Mini under Parallels Wokstation 2.1 beta 3 to see how the x264 goes. I am not sure how much performance the Parallels virtulization chews up, I will have to ask on their forums.

[izdaman]

Hey Major,

Just want to say thanks for all your dynamite work on ffmpegx, I just can't wait for the universal version, I'm so excited about it I've done a small wee twice!!

Cheers again.

[rumplestiltskin]

Major,

This question is not specific to the Intel Macs.

If I am using ffmpegX correctly, it appears that embedding a subtitle (using a ".srt" file)and enlarging the text size must be done while encoding as either XviD or Divx (don't remember which at this moment but I know that you have to be able to alter the subtitle position in order to know that the encoding will succeed).

However, this process can'e be done directly to H264. I need to embedd (and, optionally, enlarge) the subtitle during the first encoding (as Xvid or DivX) and then transcode the resulting file into H264.

At least, this was the way it had to be done up to .9v (and I had to use the .9s mencoder, IIRC).

So I guess there are two questions here:

1. Will I no longer have to swap out the .9v mencoder for the .9s to enbed & enlarge subtitles?

2. Will H264 encoding include the ability to embed and enlarge subtitles?

Thanks. This is not a huge problem.

One more thing: If this process I described will continue to take two steps, might there be some kind of a script that might be possible for me to "fill in the file names and parameters" and then run?

[major]

1. The subtitle problem is a bug in mencoder, I will report it to the applicable developers as it starts to be a little long that it is there. The latest sources seem to encode subtitles but don't take font changes into account.

2. mencoder H264 handles subtitles, not x264 H264 (however the same as 1 will apply).

[major]

Hey Major,

Just want to say thanks for all your dynamite work on ffmpegx, I just can't wait for the universal version, I'm so excited about it I've done a small wee twice!!

Cheers again.

Thank you for your support!

[major]

On a side note, it looks like in the Intel version it is possible to restore the audio level feature for mencoder.

[izdaman]

Hey Major,

Just want to say thanks for all your dynamite work on ffmpegx, I just can't wait for the universal version, I'm so excited about it I've done a small wee twice!!

Cheers again.

Thank you for your support!

No Probs. My support is waiting in the form of registering it via PayPal as soon as it's available! Not wishing to bust your balls at all but are you any closer to a release date? Thank you again.

[major]

I'm currently testing, there are a few bugs needing to be fixed. I think I'm still in my initial estimation of 6 to 146 hours (which is now upped to 99 hours).

[avramd]

Major,

Is it possible for us to drop these UB encoders into our existing installations while we're waiting for you to publish 1.0? Would they work? And/Or are you holding out on UB to finally get all the freeloaders to register? (Which I totally support)

-Avram

[mrwho]

Since ffmpgx is a GUI for the command line encoders this "should" be possible.

Ohhh, i'd kill for a native h264 now !

But if major is still counting in the ~100h range, then there're some big bugs there

[major]

It is almost done. I'm now adding a couple bonus features so as PPC users will also find something useful in the new version, 0.0.9w. Some observers believe that it will be ready for release in 27 hours, 12 minutes and 31 seconds.

[jasenko]

Great news major, to celebrate that, I'm going straight to your website to register now.

[cokeaholic]

Great news, please keep us updated.

[ultim8fury]

With todays announcement of a 17" MBP, I'm just days away from being intel only. Please please release the Intel/UB version of FFmpegX. Or just slip me a beta.

Thanks.