Tuning code for a specific system has often been necessary, but rarely enjoyable or easy.  With many variants of a given architecture now being used in machines, for example the many implementations of the x86 instruction set, tuning has become even more difficult. In this talk, I will outline our work on the Active Harmony Auto-tuning system.  I will present some results showing that we can improve the performance of real codes, and that the best configuration can vary quite a bit even among very similar CPUs.  Our tool is capable of both online (tuning for a single execution) and offline (tuning using training runs) operation.  I will also describe our core search algorithm, Parallel Rank Order.