[MPlayer-users] H.264 and advice on new hardware.

[Corey Hickey]

Mathieu Monnier wrote:
> Hi,
> 
> At the moment, the best software solution is using CoreAvc. It's by far 
> the fastest h264 decoder ( faster by 50% ), and the only one that 
> actually allows realtime decoding of anything that is not interlaced ( 
> on my computer, a amd 64 3200 + ). It's even faster than any current 
> decoder using GF acceleration ( I'm beginning to sound like an ad guy, 
> which I'm not ). Drawbacks : not available on linux (yet), not open source.
> 
> Lavc won't be able to decode high bitrate 1080p video in realtime, 
> whatever the computer. And there are still a lot of optimizations needed 
>   in order to do so ( and even so, in it's current architecture, it 
> might not be possible ). On my computer, it chokes on 1080p content as 
> soon as the bitrate gets a little too high ( one of the hardest : 
> http://trailers.nerodigital.com/Disney/TheGreatestGame_HD_AVC.mp4, 
> another one : siryana trailer from apple, where bitrate surges to 40 mbps ).


You spelled it wrong. :)

http://movies.apple.com/movies/wb/syriana/syriana_h1080p.mov

> So, in the current state, I think Windows is alas mandatory.

Perhaps for you, but for a faster CPU lavc will suffice. I can't comment 
about CoreAVC (I have heard it's fast), but for someone about to buy a 
brand new Athlon64, there shouldn't be any problem with lavc.

1. I have an overclocked Newcastle Athlon64 3400+ running at 2556 MHz, 
and it can almost, just-barely-not-quite decode those two videos in 
realtime. There are scattered scenes where the video drops behind by up 
to 0.4 seconds, but it catches back up after a moment; overall, it never 
gets left behind.

2. Every once in a while Loren Merritt makes a slight performance 
improvement to lavc's h.264 decoder. Over time, these add up.

3. Oded Shimon is working on an AAC decoder that, so far, appears to be 
performing much better than FAAD, the current AAC decoder in mplayer. 
This will benefit the Apple trailers slightly.

4. Using '-lavdopts skiploopfilter=nonref' will speed up decoding 
significantly without any readily-visible differences.


Other than that, though, I really wish we could beat CoreAVC. If I had 
any programming expertise in that field, I'd help.... :(

-Corey