I asked this on their IRC channel and Manano (I think) said that it lets the encoder skip further consideration of a block under certain circumstances, sort of like the Fast P-Skip option. He said to enable it (i.e. disable DCT Decimation) if you are using Trellis (and you probably are).
So DCT Decimation is somewhat comparable to Fast P-Skip and is advisable to use then, thanks for the quick answer :). Time to go pester Sharktooth about profile updates then ;).
------------------------------ 2000 Ferrari 360 2004 Yamaha R1 2007 Porsche 911 C2S Cabrio 2017 VW Jetta (DD) 1995 M3 (SOLD HELP ME FIND IT TO REBUY) 1998 740IL (Sold) 2003 x5 4.6is (Sold)
Nice, nice. Of course I don't expect a noticable quality change at a constant quantizer of 24, and no one switch (save for the super-important B-frame, cabac, etc) is expected to produce any easily discernable quality difference at a nice-looking constant quantizer anyways... So ya. Ya! PS: In my previous post, I was just "adding" my result to the others, but I misspelled "add" as "odd" on my Pocket PC... ;)
*bump* :p I'm reviewing my presets. We have now AQ and PSY RDO or FGO depending on the patched version we use. Before all these options I was using CQMs and/or low deblocking values. Now I don't need them, but --no-dct-decimate is still there on all my 'quality' presets. Does it make any sense to keep this option if I'm already using AQ and PSY RDO (or FGO)?
------------------------------ LowTech 2001 740iL E38 Loaded 2002 745Li E65 Loaded Link to LowTech's Dual Sevens
If you're using PsyRDO, it's advisable to use full trellis. And if you're using full trellis, I've heard it said before that --no-dct-decimate is a good idea. I believe it also helps a little with grain retention, but the difference is small. I definitely wouldn't use it on anime/cartoons, though.