I have, core 3 from my quad is noticable and repeatedly slower then the other 3 cores, though that could be the ihs not being perfectly flat since that core is usually hotter ( hotter=slower? ).

Maybe Safan was saying it would be fun to be able to target specific hw ( say you know the nv architecture and want to bench one set of stream processors directly ( nvidia groups sp's into groups of 8 iirc? )).. That way you can find out if any of the shader groups is faster or slower then the other. Though I don't think you can do this with CUDA.