Floats faster than doubles here!

[Psychophanta]

Not needed high res timer for this. I made a hires timer, so show us yours.
Beware! Hires timers can lie, because they are based on a time unit (a minimum internal amount of time) which is taken as time unit by the the complete hardware and there can not be nothing lower than it. The problem is that that amount of time is about some thousands of nano-secs, which is a lot of time.

Yes, in general i think Intel has lost the battle, at least for now. Just see the Comtois results, they are a shame in front of a Athlon64, and even AthlonXP @1800 MHz for example

[dagcrack]

Use a hardware hires timer then!
Sorry but I don't share much over here (got my own reasons).
You'll need a hires timer for anything you're benchmarking, more if its a function. A lower resolution timer would lie in that case not a higher one. Plus you are not dealing with NASA's shuttles here, so, put that perfectionism away from my sight. Seriously though. You have a higher resolution. What is the point about talking if this higher resolution is due to be trusted or not? when indeed it is, you might just have a small discrepancy but nothing to be compared with those 10ms discrepancy you were dealing with by using GetTickCount();


Also.. before you start a test, have a 2 seconds delay or so, and make sure you're wasting the "same" amount of cycles during your tests (say you closed an application that was using about 30% of your processor and now you'll think your function is faster?)... Not to forget about this other tip: use as much iterations as possible and make as many tests as you can per try. for example, 10 tests at 1000000 iterations, then you get the average of the results. this will be a much better number to work with and compare from later on, than a single result number... That way you can iron the discrepancy down.