as far as i know K33 and K34 plots under madmax use some precalculated data but not for k32 maybe this is the problem, but this is totaly speculation from me.
on the other hand it happens only at the 2nd part (the rewrite part of the tabel) of every taple except tabel 7 in phase 2. This part is very I/O intensive (SSD or RamDrive) and the cpu is not fully occupied at this time. so maybe some coding in this part is buggy or not optimal.
Maybe you can post about it on MM github discussion page - Discussions · madMAx43v3r/chia-plotter · GitHub
Max is working on his coin, and plotter speed / predictability may be also important for him to check this problem out. I would imagine that it possible, he would like to have his plotter to be as close if not better than BB (or maybe he is already working on MM2).
Otherwise, without a more verbose output, or potentially running it with a debugger, it is just a hard problem to crack.
1 Like
I’ll give it a go over the weekend, thanks.