TAK 1.0 - Beta testing Reply #50 – 2007-01-06 16:44:25 Quote from: Caroliano on 2007-01-06 15:24:43I found an very interesting file, that answers in an almost linear way to the increase of compression time, in case of TAK and MAC, and in others it is an jump in compression in the very high setings:Very interesting file! I have downloaded it and played with it.Quote from: Caroliano on 2007-01-06 15:24:43The compression delta for TAK is 4.65 %, wich seems very high for TAK. And the compression delta between Flac -0 and this extreme mode in OptimFrog is 12.04 %. Flac is the "left zero", with an delta of only 2.55, but Flake got the trick somewere beteween -8 and -12.Usually there are 2 possible reasons for such a big delta:1) The file likes the PreFilter.2) The file likes many predictors.This time the answer is 2).Quote from: Caroliano on 2007-01-06 15:24:43Another thing intersting in this file is the "max" swich in TAK. Here the gains in compression with max for each preset:Code: [Select]Turbo Max over Turbo 0.64 %Fast Max over Fast 0.55 %Normal Max over Normal 0.20 %High Max over High 0.35 %Extra Max over Extra 0.46 %It increases after "normal"! The High Max compress beter than extra simple!You are right: Usually the "max" switch isn't doing much for presets higher than FAST, because starting with NORMAL most of the options "max" could add are already part of the presets default setting.The option "Frame partition calculator - Validate" is slow and quite useless (on average less than 0.05 percent advantage) most of the time. But it can help some problem files (especially with high predictor orders) and this is one of them:Extra + Max: 67.77Extra + Max without Validate: 68.16Difference: 0.39There are some more hidden options in TAK, which can help rare problem files. Currently they are disabled, because they are very slow and on average have as little effect on compression as Validate.Thanks for providing such an interesting sample for my test corpus! It can help me with further optimizations.