PDA

View Full Version : Scaling of SMP folding client


edcentric
30 Mar 2007, 2:30pm
Has anyone tested to see how folding scales with increase in cores vs increased speed?
I assume that the folding speed will scale directly with CPU speed, but what about with the number of cores? With the old method it is linear, but does SMP change all of this? Does going from 2 cores to 4 more than double the folding speed? 8? 16?

I am thinking of getting rid of a number of boxes, and building a single workstation. If I could get more folding out put it might be worth it.

EyesOnly
30 Mar 2007, 9:20pm
One thing i know is that going smp will really boost your production. But that's now what you asked.

Leonardo
30 Mar 2007, 9:55pm
I assume that the folding speed will scale directly with CPU speedThat assumption is correct. I don't know about number of cores above two.

Before you remove old computers from service, I would watch SMP for a month or more so. It remains to be seen what the point structure will be once it's no longer a beta. Mainstream processing usually ends up with mainstream points awards.

airbornflght
2 Apr 2007, 4:45am
What would be awesome if you could literally set up a fold farm, like people set up render farms.

Make on pc the router or the pieces and then all the other pc's work for that pc. Then if you had 8 quadcore machines you'd have 32 cores churning away at the same unit.

SPIKE09
6 Apr 2007, 1:58pm
Bit of an apples and oranges comparison here but my e6400 does 40 minute frametimes on the P2610 and 20 mins on the others, my pal TrickyD has the Q6600 ~ 8 mins frametimes on all the available SMP wu's

Gargoyle
6 Apr 2007, 2:58pm
The current SMP client only runs four cores, so if you had more than that, you'd probably need to run more than one instance of the client. Not sure if it allows that, though.

SPIKE09
6 Apr 2007, 5:28pm
It does on a dual quad mac pro they advocate 2 instances of the mac SMP client