PDA

View Full Version : F@H and significance of ocing


manbush
09-12-06, 06:51 AM
How much difference would it be to run F@H on a E6400 that has been overclocked from 2.13ghz to say...3ghz in terms of time and WU's completed etc?

Edit: Is it also okay to run two F@H on my development machine? I don't want any slow downs or anything in compile time or would running 2 F@H actually bring down the WU's complete?

WarriorII
09-12-06, 08:19 AM
ChasR would be the most knowledgable to answer you question.

but, I just wanted to say, that OC'n is a Good thing. :)

It's free speed for your pc.

And to "quote" the little green lizard:

"Free is always good. Everybody wants "Free". "

Fold On !
:attn:

||Console||
09-12-06, 08:40 AM
Only way you will know the difference is run @ stock do some benchmarks , then oc it and test it again

manbush
09-12-06, 08:58 AM
Here are my folding rigs:

E6400 2.13ghz
AMD 3700+ 2ghz
AMD 2000+ 1.6ghz
P4 3ghz HT
P4 3ghz HT

1 is dual core, 2 are hyper threaded.

I can overclock all of them and gain atleast 30-100% OC gain. My number 1 worry is my electric bill.

The 2 P4's I dont have to worry about cause they are dedicated machines and I dont have to worry about electric bill.

How much do you think it may cost?

dwschoon
09-12-06, 11:24 AM
Usually the folding performance of a particular project increases linearly with clock speed(within the same cpu architecture.) You can get an estimate of your performance by taking your current ppd and dividing that by your clock speed(ghz) to get your ppd/ghz. Then take that number and multiply it by the speed you want to attain, and that will show you what you should be producing. Here is an exmple If you have a 2.13 conroe that gets 300ppd.

300/2.13=~140.85ppd/ghz
140.85*3=422.55ppd

So in this example, o/c from 2.13ghz to 3ghz increases production by 122.55ppd. There will be some variation due to memory timings, etc. Remember that this only holds true when comparing results within a particular project, since each project performs differently. Also, different cpu architectures will yield different ppd/ghz even on the same wu.

ChaseR keeps a database of what ppd/ghz you should expect for differenc cpu's on different projects.

edit: For the most part, it looks like you have cpu's that dont use much power. You really shouldnt expect more than a few $ more on your electric bill. You really have to have a lot of machines before the numbers get too high.

ChasR
09-12-06, 11:40 AM
In almost every case, if you OC a machine 10% you will increase points production 10%. If you have to loosen memory timings to achieve the overclock the performance will be slightly less linear. Most WUs will not be affected by memory dividers as long as available bandwidth is high enough and the performance remains cpu bound.

There are some high bandwidth WUs where overclocking's effects are even better than linear. I recall Silver's experiment with Celeron Ds where he got a 100% performance boost from a 40% OC on one particular type of WU. QMDs and p147x/81 are the only WUs I've known to exhibit nonlinear response to OCing. On these the processor was limited by ram bandwidth and the roughly 40% increase in bandwidth that came with the OC removed the bottlenck.

pscout
09-12-06, 07:36 PM
Increase in OC and folding ppd is not linear with power consumption in my experience. This is only tru up to the point were more OC'ing takes significant bumps in voltages to be stable. I have few/none that i run at stock VC.

Most Conroes and the most recent stepping pentium D's will achieve pretty high OC's without signiificant increases in volts. Increasing fsb will also increase your power draw even without increasing cpu volts. I have not seen a measurable effect of increased power draw through bumping v ddr, v mch, vfsb, etc. These comments are all based on my intel experience.

I use a ups with monitoring software to get a rough idea of power draw ... it reports in 6W increments. What i now realize tho is that i didn't measure any rigs fully loaded folding at stock speeds :o. I guess i am such a points junkie that i didn't think the measurment was relevant :D I did measure them OC'ed at idle. My main purpose in measuring them was to determine how many circuits i needed to run the folding farm.

With the doubly high costs in the hot months of both running the farm and the AC, i downclocked a few of the hottest running rigs and reduced VC. I also replaced some older smithfields and p4 ht's with newer stepping presslers to help contain electric costs without impacting production.
And now conroes not only improve production but also reduce the electric and cooling bill.