I actually have no idea when he started using the tweak (can't ask either has he ignores my pm's ), I can only tell what I've read at the Coolaler boards translated online (so crappy english).

Can't find the pic atm, but the superpi 32M run was divided in three parts, two of them which can only be slightly faster with tweaks, the last one (bonus round = loop 20 to 24) can be a lot faster using the CDT tweak and copy-waza.

Does anyone actually know how superpi is run by the system? At what point it uses the memory/cache/virtual memory?

@ Beardy: lower timings =/= faster. Subtimings have to be set at the correct level in order to be really efficient. Imho, a lot depends on how you set the CL and tRAS and the timings of the latency between both.