27 Jan
2010
27 Jan
'10
6:58 a.m.
Dan Gahlinger wrote:
but still, that sort of trick isn't seen today.
even the manufacturers of cpus have bought into the "faster mhz is better"
trick,
rather than multiprocessing.
Ultimately that depends on what you're doing. If you're doing something
that can be easily split across multiple CPUs (e.g. matrix maths,
raytracing, some types of image processing), then a multi-core CPU beats
an uber-GHz CPU almost every time (or at least has the potential to).
Best case, you're looking at around (cpu_speed * number_of_cores)
equivalent on a single-core CPU (less a bit for the threading code).
If you're single-threading then your quad-core 3.2GHz uber-CPU is only
going to work as well as a single-core chip of the same speed (although
if the OS has a multithreaded kernel, it might push some of the grunt
work of OS management / HW interfacing onto a spare core).
Take a look at some of the fairly new "GPGPU" concepts -- nVidia's CUDA
for example. A typical GTX200 class GeForce chip has ~192 cores running
at 1.7GHz, and can easily run rings round most modern CPUs (within
certain limitations). It's mainly intended for e.g. running the same (or
a very similar) operation on a ton of input data in one go -- things
like raytracing and games-style (particle engine, visual effects, etc.)
algorithms tend to work well, as do brute-force password-cracking
engines (!), as proven by the likes of Elcomsoft...
--
Phil.
classiccmp at philpem.me.uk
http://www.philpem.me.uk/