[MEncoder-users] Tuning hqdn3d

[vmrsss]

On 10 May 2009, at 20:23, Phil Ehrens wrote:
> 2-pass ALWAYS produces better quality. Visibly better,
> not just PSNR hand-wavey better.

Ahem, this is grossly exaggerated, the difference between 1-pass CRF  
and 2-pass at the same avg bitrates is negligible. There are tons of  
threads on issues like this in the X264 forum at doom9.org. Just  
search. The actual problem is that with CRF there is no way to predict  
the avg quantizer, and therefore the final size.

> With the way x264 handles threading, I wouldn't be surprised if
> 2-pass+turbo mode was just as fast as CRF, on systems with several
> cores.

Of course it is not, the turbo pass may be fast, but you still have to  
run it... Possibly including prefilters (scalers, deblocker, deringer,  
denoisers, sharpeners,..)

> Take a look at all the videos you've encoded with the same CRF  
> settings,
> and I'm willing to bet the vast majority will have a pretty similar
> bitrate.

Experience shows that that's not the case. It's the entire point!

One thing interested me in this thread, and it's the assertion that  
hqdn3d/nr is useless for one pass (?? yet, I can't find anymore). Why  
would that be so?


PS.

> Maybe when I have a CPU in the 6-7 GHz range.

This ain't likely gonna happen, apparently they start to melt at about  
5 GHz.