Improve speed on H264/AVC video conversion

[klode]

Hi,
First of all, I want to thanks all people who in this Forum found solutions on billions of issues/trouble/questions and much more!

I have a machine with Intel QuadCore Q6600, 4GB Ram 800MHz, and an ATI HD4550. I want to improve speed on H264/AVC Video Conversion.

Actually, I'm using MeGUI with ffdshow as codec packet. I get good results, but I don't have good speed as supposed. For example, if I want to make a 720p video conversion, I get 30/35fps on analysis step, and 18-19 on conversion step.

If I want to convert a 1080p, I get 24-25 on analysis step, and 9-10fps on conversion step.

NOW: can I use the graphic card for conversion? And how can I do that? Or maybe does I need to change graphic card with something with CUDA technology?

I wish you can help me with your great Xperience!

[ozok]

Well, using 64bit versions of tools -whenever possible- makes a difference.

I don't know if your CPU supports it but I guess you can replace -at least- x264.exe with 64bit one if you haven't done that already. I don't use MeGUI so I don't know which tools in its process chain can be replaced with 64bit versions.

An other way is to use a faster preset.

[deadrats]

there's really only one way to get more speed and that's to upgrade, everything else is just incremental:

http://legitreviews.com/article/2247/1/

[jagabo]

can I use the graphic card for conversion?

Yes, but not with MeGUI.

And how can I do that?

Use software that supports GPU encoding. Search and you'll find several.

Or maybe does I need to change graphic card with something with CUDA technology?

ATI's GPU h.264 encoder is pretty poor. Nvidia's is better. Both are poor compared to CPU encoding with x264. And x264 with fast settings is faster than both of them -- and still delivers better quality. Intel's QuickSync encoder is faster than x264 at fast settings but still delivers lower quality. Of course, you'll need a new computer to use that.


Not much has changed since these reviews:
http://www.behardware.com/articles/828-1/h-264-encoding-cpu-vs-gpu-nvidia-cuda-amd-str...-and-x264.html
http://www.extremetech.com/computing/128681-the-wretched-state-of-gpu-transcoding
http://www.anandtech.com/show/5835/testing-opencl-accelerated-handbrakex264-with-amds-trinity-apu

QuickSync has gotten faster with Ivy Bridge, and again with Haswell. Quality hasn't changed much.

[hello_hello]

Actually, I'm using MeGUI with ffdshow as codec packet. I get good results, but I don't have good speed as supposed. For example, if I want to make a 720p video conversion, I get 30/35fps on analysis step, and 18-19 on conversion step.

If I want to convert a 1080p, I get 24-25 on analysis step, and 9-10fps on conversion step.

Are you referring to 2 pass encoding when you refer to analysis and conversion? ie first and second pass? If so, do you need to use 2 pass encoding? CRF encoding doesn't run a first pass so you save all that time, but the actual encoding process happens at pretty much the same speed either way.

If you're actually running an analysis pass, fee free to ignore this post.

[klode]

If I use a faster preset, does I have a good quality anyway? What are the difference between a faster preset and a "accurate" preset?

Thanks jagabo for your threads

hello_hello: yes, I use 2 pass encoding, I think for a better solution on re-encoding some AVC video.

[hello_hello]

2 pass or CRF, if the encoded video has the same average bitrate it'll be encoded the same way. It's not like the old days, Xvid's single pass quality mode is a kind of "dumb" encoding mode. x264 has the same quality based encoding method as it does for 2 pass.

Of course if you need a particular file size you'd use 2 pass and the quality will be the unknown, but as hard drive and disc space doesn't cost a fortune as it did back in the "Xvid/AVI fit a movie on a CD" days, it's more common (around here) to pick the quality and let the file size be what it needs to be. After a while you'll find a CRF value and resolution etc which gives you... on average.... a file size you're happy with.

The same applies when re-encoding AVC. The encoder doesn't know how the original video was compressed or the bitrate used etc, it just gets fed the decoded video which it re-compresses. So to ensure the quality is pretty much the same as the original you'd probably use CRF 18, or maybe go down to CRF 16 if you want perfection.

I've never used a preset faster than the default of medium. I think the speed presets can effect the quality for a given CRF value, but it probably won't be noticeable until you get to the extremes. Slower speed presets should reduce the file size for a given CRF value (the video might be compressed more efficiently) but that's not always the case. If you encode using a particular file size then in theory at least, the faster the preset the more potential for lower quality as the video mightn't be compressed as efficiently.

[jagabo]

If I use a faster preset, does I have a good quality anyway? What are the difference between a faster preset and a "accurate" preset?

The difference in quality between x264's "veryfast" preset and its default "medium" preset is not huge. If you look very closely you'll see some moving edges get a little jagged and a little less grain is retained (and hence there will be a little more posterization, especially in dark grainy areas). The presets built into x264 are:

ultrafast, superfast, veryfast, faster, fast, medium, slow, veryslow, placebo

When using CRF (constant quality) encoding the general rule is: the slower the preset the more compression you get. But the "veryfast" preset is a bit unusual in that it usually gives more compression than a few of the higher presets, up to about "slow". File sizes get much larger at superfast and ultrafast (as much as 300 percent larger) but the difference between veryfast through placebo is in the range of 10 percent.

I posted a list of individual x264 settings set by the presets:

https://forum.videohelp.com/threads/347067-VirtualDub-H-264-Encoder-Speed?p=2169830&vie...=1#post2169830

I don't know how these presets fit in with MeGui's naming conventions.

[klode]

Thanks a lot for all of us. My goal is to compress video at defined size, so I don't think I can use CRF, because it could be make a file bigger than dimension I need. So, for a defined dimension, can I use 2pass enconding or this is it useless?

[hello_hello]

If you encoded the same video twice, once using CRF and then a second time using 2 pass, and the average bitrate for each encode happened to be the same, both would be encoded in almost exactly the same way.

So the difference is CRF lets you pick the quality while file size is unknown, and for 2 pass you pick the file size but the quality is unknown.