Quote from Jack_Larkin:
Hey, that's cool.. looks like the developers of that app focused solely on the 64bit architecture and hardware...
Which app is this?
The 64bit OS plus 32bit binary should be the same as the 32-32 setup.. so that's an odd result. There isn't any emulation going on since the architecture is nearly the same (64bit is mostly just a memory registry extension.)
Anyhoo.. one key thing is how developers code for target hardware. Sometimes they code specifically for newer chips (which pretty much all new processors are x86_64 capable now) and utilize any hardware functions that could speed up specific types of computations. To then take this code and make it 32bit friendly from a compatibility point of view, they often cut all hardware specific optimization down to just "i586/i686" base compatibility (think back to Pentium 3s). This pretty much kills performance when using it in 32bit mode since most newer hardware assisted computing features are left untouched. So a lot of what you're seeing can be related to how the developers put out their software.
Think about it, how would the same hardware do the same task that much slower when the only difference is it has less addressable memory to work with at the time of processing yet it hardly breached the less addressable memory limit anyway?
....this isn't me downplaying your results, just giving more info on what could cause them, since a 100% difference isn't normal.. but I can't fault you for presenting and arguing for what you've experienced first hand.
If you'd like, I can point out a few places online that collect benchmarks of various application types and workload types and test them in different software/hardware setups... this would better illustrate my point.
