[MEncoder-users] Tuning hqdn3d

[vmrsss]

On 12 May 2009, at 04:50, RC wrote:
>> Experience shows that that's not the case. It's the entire point!
>
> Again, you're welcome to post some numbers.  At best, your assertions
> are highly subjective.


Indeed, highly subjective as are yours. And in fact, I said  
"experience shows". I don't keep benchmarks of what I do, less than  
ever on different hardware. Do you? Nevertheless, I maintain that it  
is just enough to encode a TV series to observe significant  
fluctuations of the avg QP for a fixed CRF by the episode.

> The 1st pass is not just a throwaway step.  It takes care of most of
> the processing that can't be multi-threaded, allowing the second  
> pass to
> much fully utilize multiple cores.


Yes, this is partially true, but we are still far from saying that 2- 
pass is as fast as 1-pass. And you keep discounting all preprocessing  
which may be needed in both passes.