Log in

View Full Version : megui with x.264 64bit


daWsOn_s
20th July 2009, 01:29
Hello,
is it possible to use megui 32bit with x264 64bit? I read that it could improve enconding speed and i'd like to try. I installed avisynth x64 with x64 plugins but enconding won't start.
There's a post here with a x64 megui build but the link doesn't work.

Any help and experiences?

Thanks

Rain724
23rd July 2009, 00:22
I'm having the same problem. It looks like MeGUI can't launch the 64bit x264 process to start the encoding... Any help/information about this would be greatly appropriated.

Kurtnoise
24th July 2009, 09:48
to use this correctly (megui 32bits & x264 64bits), we need to tweak the encoding pattern used (i.e add avs2yuv in the chain)...

Otherwise, that would be to compile megui and all its components as 64bits. However, as far I can tell, an internal component is not compatible with this. So, it requires some rewrite code.

The best way at the current time, it's to use all 32bits apps/tools in your favorite 64bits OS..

alwa
24th July 2009, 11:05
If you use 32-bit Avisynth you could be interested in this (http://www.mediafire.com/?3wykjxjjine) (basically an extraction of the AVIS-input of x264). The program redirects the all command line parameters and pipes the video to the 64-bit executable of x264. Just place the x264_x64_wrapper.exe and the x264_x64.exe (any 64-bit executable of x264 named like that will do the trick) in the same folder. Then reference megui to the x264_x64_wrapper.exe.

benwaggoner
26th July 2009, 02:49
Interesting.

Has anyone found a combination where the "thunking" overhead (or whatever you call it) is small enough that the perf adavantage of 64-bit still makes it worth the trouble?

twist3d
26th July 2009, 08:16
Interesting.

Has anyone found a combination where the "thunking" overhead (or whatever you call it) is small enough that the perf adavantage of 64-bit still makes it worth the trouble?

lots of benchmark results (including mine) at http://forum.doom9.org/showthread.php?t=144140.
I think when we get fully chained 64-bit encoding process, the speed increase will be more noticeable.

daWsOn_s
29th July 2009, 20:52
If you use 32-bit Avisynth you could be interested in this (http://www.mediafire.com/?3wykjxjjine) (basically an extraction of the AVIS-input of x264). The program redirects the all command line parameters and pipes the video to the 64-bit executable of x264. Just place the x264_x64_wrapper.exe and the x264_x64.exe (any 64-bit executable of x264 named like that will do the trick) in the same folder. Then reference megui to the x264_x64_wrapper.exe.

It works, but the speed is the same :( x64 versione is a little bit slower too :confused:

turbojet
1st August 2009, 08:47
Thanks for the wrapper aiwa. It's about 5% faster then comparable x86 builds for me on an X2 3800 and Q6600 both Windows 7 RC. It's not quite as much of a gain as I see with pipebuf + avs2yuv which is around 10% but I'll take any speedup at same quality. Could it be number of buffers used?

alwa
1st August 2009, 15:27
It works, but the speed is the same :( x64 versione is a little bit slower too :confused:

It happens for some encodes. You can see this behaviour also in some the benchmarks of: http://forum.doom9.org/showthread.php?t=148431
But with the slow preset i got always reasonable better results.


Thanks for the wrapper aiwa. It's about 5% faster then comparable x86 builds for me on an X2 3800 and Q6600 both Windows 7 RC. It's not quite as much of a gain as I see with pipebuf + avs2yuv which is around 10% but I'll take any speedup at same quality. Could it be number of buffers used?

The buffer is not adjustable (yet). Mainly because i doubt there would be a big improvement since no "always good" buffersize has been discovered yet. That it is overall slower than the pipebuf & avs2yuv chain is weird. What buffer values do you to get the 5% better results with pipebuf & avs2yuv

turbojet
4th August 2009, 04:36
Depending on settings/cpu 2 and 4 MB buffers were the fastest, sometimes considerably faster (5+%) than the other buffer sizes.

Maybe 3 MB buffer would be optimal? it hasn't been tested yet.