The min-phase/linear-phase filters have the same computational requirements–same filtering code + same filter order.
5th/7th order makes a very small difference, but the difference is mostly lost in the superscalar characteristics of the CPU (the 5th order modulator can’t fully utilize the capabilities). Memory performance is also significant for the SDMs.
1.2 and 1.3 are essentially within measurement error, especially if your CPU has frequency scaling turned on. The measurement is a simple “time it took to process audio / amount of audio processed”, so if the 3GHz CPU is running at 2GHz it’s going to read differently.
Since my last post, we built support (not yet released) for splitting up work across 2 cores, which (combined with the aforementioned performance work) makes DSD512 practical on a NUC 5i5’s and DSD256 practical on NUC 5i3’s without compromising quality at all.
(This signal path was captured on a NUC 5i5 running a pre-release of ROCK. I’m not sure I’d recommend the i5 for DSD512, but it’s possible, and I’m able to live with it).