His implementation does a little better than the ethminer (genoil or otherwise) ones when it comes to speed on the compute side - i.e. the computations. He also benefits from a technique I also use - somewhat poorly implemented on his side, yet likely somewhat effective - that is, executing some operations "for free" because the memory read is being done at the same time. This also increases the speed relative to the other public miners, and the side effect is (of course) the ASIC is idling less of the time... resulting in your raised power consumption. Appears to be touched up output from the AMD compiler ( well, almost certainly, seeing as it left the little love letter "@(#) OpenCL 2.0 AMD-APP (2442.12). Driver version: 2442.12" as well as another noting it was used with the compile option "-legacy" 
)
)ohh is that what's he's doing, no wonder it seemingly has very hard to find sweet spots
profiling measurements and parallel processing through asm optimizations, under loads is very effective, yet less scalable and stable under various conditions than 'clean' implementation. - proper (not fine tuned for specific results asm optimizations that can benefit from compiler targets optimizations as well. but then he uses his own custom environment ...EDIT: looks like on various timings and clocks [increasing the core clock then should be generally good solution for 24mh drops] it can also produce "interlaced like" fragmented memory load [mem read is done, but the filled in stuff isn't:)] and thus power draw inconsistencies. will check though with adding more RAM, so maybe not. AMD blockchain and new API is not very nice. Is there an update we should expect from him in a few weeks?
and i don't see my coins on etermine, what can i do? thanks
