[sdiy] SPICE and Processors

[Samppa Tolvanen]

Aww, So after 10 years of the first SMP capable entry level x86
systems emerging, the situation hasn't changed for better. Oh Joy!

On Sat, Jul 25, 2009 at 7:51 PM, Rainer Buchty<rainer at buchty.net> wrote:
> On Sat, 25 Jul 2009, Neil Johnson wrote:

>
>> As a quick example, running a phaser simulation in version 3 of LTSpice
>> with audio in and out I find a Core 2 Duo 2GHz 2GB machine about 20% faster
>> than a single-core Pentium-M 2GHz 1GB machine (both Dell laptops, WinXP).
>
> Which might well be related to cache and general task-switching issues
> leading to an overall more balanced execution of all running tasks and
> somewhat less L1$ clobbering, therefore having nothing to do with the
> potentially doubled processing power per se.
>

Well. You put it that way. But even OS performing house keeping uses
resources of its own: What You see in Task Monitor is one core taken
up to the full load and the other under some. "Win" without
deliberately told so, schedules processes with such priority, that GUI
is kept responsive.

In uniprocessor systems multithreading would mean a performance hit of
task switching and depending on programming style, when threads need
synchronization.

On these New-Breed 64b systems rewriting for multiprocessor
environment should be seriously considered. Even on x86 properly
paralled application hit close to "times CPU" performance on SMP.

Samppa