On Thursday 09 February 2006 20:48, [EMAIL PROTECTED] wrote: > I have a 3.0GHz P4-D (dual-core) CPU using XP Pro > SP2... > > I'm presuming you could run two instances of Prime95, > one on each core of the CPU, as if it was an dual CPU > system... > > I'm also presuming this would be more efficient than > even a hyper-threaded CPU...
Definitely. > > I'm also presuming unless you have a program > specifically designed to use both cores or able to be > assigned to a specific core, Windows by default will > use only one, ignoring the other completely... > > Are these presumptions correct? No. AFAIK if a process/thread isn't specifically bound to a particular core, it will execute on whichever becomes free next. The linux kernel scheduler modifies this a bit by keeping the process/thread on the same core it used last time unless the wait would be unreasonably long - this saves a certain amount of load due to cache invalidation, refetching registers etc. > > If so, I'd like to assign an L-L test to the ignored > core, and do a trial-factoring on the other one > (presuming this would be the most efficient use)... I suspect this would be optimal. > I'm presuming the ignored core, would be core 1 (or in > Prime95's case -- CPU 1)??? Ummm - a dual core CPU is supposedly the same as two single core CPUs except for the packaging. My best guess (based on experience with dual CPU SMP systems) would be to assign factoring to CPU 0 and LL testing to CPU 1. The point being that there seems to be a built-in bias to using CPU 0 for interrupts in Windows (at any rate in Win 2000) and factoring recovers faster (having very little in the way of cache contents to lose). Regards Brian Beesley _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
