Dual Core Systems

Right I’ve got my core duo running 2x folding nicley now. Did have one core folding and one core Riesel Sieve LLR searching but when I was running that, the LLR was EXTREMELY slow.

And after some topics on the folding forums, it says that some folding units are cache dependant (more show than others) and thus when running a dual core the combination of the two units can effect one another.

Now this can obviously affect the PPD of each unit and lowers the PPD. Now the question is, is it more benificail to run 2 cores or just the one?

I was thinking that 2 cores will lower the PPD of each unit running but you’ll actually get a higher output from the PC of points in the long run. Am I right in this assumption?

I know there’s a couple of you on dual cores out there so I thought I’d ask.
And seeing as people says it’s getting easier, people like me still need a hand evernow and then

if you’re after points, run two instances, science likes the results returned quicker and so if you’re in it for hte science run one.

An old rule from Folding stuck in my head :slight_smile:


Thats what I thought it was about. I’m in it for the both really.

I’ll stick with the dual core running I think. I mean the slow down isn’t that major I think. I have a WU benchmarking at 36 minutes and my main rig benchmarks at 33 minutes for that unit so taking into account that this isn’t an AMD 3500+, I think it’s not entirely noticable. It might be because I have one configured for big WU’s and the other for small.

Pentium M and Core don’t do well at LLR, on a “per clock” basis, and they don’t have much of that either. Personally I wouldn’t run LLR on it. It is better suited to sieving if you want to stick within the rieselsieve group.

This certainly doesn’t apply to Core 2 which absolutely wipes the floor with any other x86 CPU, per clock, at LLR.

In general with multi CPU architectures, you get more total throughput with 2 instances than one. This includes even the P4 HT. However the gains do vary depending on the limiting factor so at times the total throughput may be barely over that of a single instance.

In cases like these you may want to choose different projects per CPU to maximise on usage. If you need really small footprint projects, I know that RC5 and OGR from distributed.net are, either of which can work substantially out of the CPU cache and therefore not choke memory bandwidth or other resources.