Looking at the opencl kernel, can the belowOrEquals function not avoid the endian related comparison of seperate bytes, instead moving the switch into the python code when creating targetH and targetG. Then less branches in kernel and perhaps better stream usage?

I've not spotted how to figure out what hash speed the entire network is searching at? Can anyone link me?

I get this problem (Ubuntu 10.4, bitcoin 0.3.19 built from src). Not too keen on having to override my font preferences to avoid it.

Will try and debug at some point.