Re: [tlug] 64bit is faster than 32bit

[Joe Larabell <fred62@example.com>]

> An author of the graphic library GraphicsMagick says[1] 64 bit AMD is 
> 30% faster than 32 bit. Is that true?

Ya first gotta ask: "Faster for what".

If they're doing lots of high-precision math, I can believe 64-bit could 
have an edge. The product I work on at my day job has been benchmarked on 
32-bit and 64-bit machines and the 64-bit version is slower. The reason, 
according to our in-house experts, is that the larger word size makes 
cache misses more frequent (for the same size cache, of course). There's 
probably also a penalty for loading the executable (since all pointers 
will take up twice the space) but I've never heard anyone claim that as
a significant factor. Raw performance is such a hot button in my field 
that I'm pretty sure the code generation gurus would have squeezed more 
performance out of the 64-bit machines if there was any there to be had.

In the case of GraphicsMagick, this guy claims that the application has 
it's own swapping mechanism that kicks in when the memory runs out. If you 
hit that point, the performance will go to hell no matter how many bits 
you're using. The fact that the point where you exceed the address space 
is occurs much later, if at all, on a 64-bit machine will obviously affect 
the total performance but that's not because the 64-bit processor is any 
faster -- it's because the disk is orders of magnitude slower.

Of course, the original argument assumes you have more than 3GB available 
on the machine. Othwerwise, the O/S hits its limit and starts shuffling 
stuff off to disk anyway, and that's irrespective of the bit-width of the 
CPU.

---
Joseph L (Joe) Larabell            Never fight with a dragon
http://larabell.org/                    for thou art crunchy
http://thelemicleague.org/        and goest well with cheese.