Is there a point to keeping the first pass?

[NuclearFusi0n]

as opposed to "discard first pass"?

[sam_b]

Curiosity mainly. I think. And there's sometimes the possibility that the first pass size is just right, saving the need for a second pass.

[hayami]

same here...=P
Sometimes the 1pass just fit 1 cd so I'll just burn it right the way.
However, writing the 1st pass file may takes a little bit of time instead of discarding the 1st pass.

[TheXung]

If you do an extremely long filter chain while encoding the first pass, saving it and then doing the second pass on the saved first pass, it saves you the trouble of doing the filter chain for second pass. Though, this could come at a quality loss.

[ErMaC]

Xung that would also make your previous stats file almost worthless because you're now encoding from a different stream than the stats file was made for! Even though everything may line up, the added DCT processes and compression (especially if you're using bframes) will make your final encode far worse.

If you want to save the time of doing your filter chain over again, encode to huffyuv and then compress from.

Or wait until -h finishes his lossless YV12 compression codec and use that with AVISynth 2.5 to avoid any color conversions at all!

[TheXung]

you're now encoding from a different stream than the stats file was made for!

What do you think those stats are of? They're framesizes of the movie at quant 2, which is also the exact same thing that's being outputted.

Even if somehow the 2nd pass is just slightly different, what difference does that make? The whole point of a 2-pass encode is so you know the shape of the curve and can compress it at desired spots to desired degrees. Even if the second stream is slightly different, the curve will still have the same shape; it's not like you just rearranged the fast and slow motion scenes. Also keep in mind that frames have to be quantized to distinct integers, it has to be either 2, 3, 4, . . . You can't have a quant of 3.6.

[sam_b]

@The Xung

Yes, it does represent the quant2 sizes of *the source*. A second quantisation stage will obviously produce a smaller output than the first, rendering the stats data invalid. And it will alter the curve to some extent as many of the HF coefficients will have been removed first time round. Depending on how many of these co-effs were there in the first place, a re-compresion will have different effects. A noisy/sharp scene will compress differently to a smooth/soft scene second time round.