The one big difference is that open source miners have excellent cpu code for the algo, borrowed from their cpu miner. For rigs with slow celerons, we will add significant latency with our more naive on-cpu verification. We need to implement automatic generation of x86 machine code for the random ops in CN/r, quite annoying crap to spend time on, but we don’t really have a choice it seems.

Fix for now: if your rig(s) are running without any hw errors, try running with —no_cpu_check. This will disable our check and let the pool run the check instead. Will be fine on most pools and if you don’t oc so much that you get invalid shares that you flood the pool with .