That's quite true, and that qualifies my reply, too. I responded under the assumption that variations in runtime (due to variations in rated clock rate, cache size, etc.) were assumed to be differences that weren't counted as differences in result. Just to illustrate the assumptions under which I framed my response, by taking those assumptions to a hypothetical extreme, it's not fair to penalize a single 80386, with 4GiB of RAM connected to a humongous disk executing some AI algorithm, for the long runtime it takes to calculate a result, if the objective for such a system is to calculate only a single intelligent (or perhaps "intelligent") response according to some theoretical criteria (e.g., compute a theoretically optimal turn in a game of Civilization, no matter how long it takes to figure that out). If the objective is to generate a timely response in realtime under dynamic conditions (that meets certain criteria of effectiveness), your hardware would be radically different than that, and you wouldn't be speaking just of calculations, as if that was all that mattered.