Optimal folding?

[PTRMAN]

What is the optimal setup to fold?

I'm currently using 1 GPU and 1 SMP client. My Q6600 is not OCed, but I've bumped up my 8800GTX by a tiny bit using EVGA's Precision. SMP WUs usually take about 25-30 hours, while a GPU WU takes about 2.5 hours.

I'd like to explore OCing the Q6600 - I've read the guide and am studying the BIOS. One of these days I'll take the plunge and give it a whirl. Presumably that would help with the SMP units.

There are obviously a lot of different F@H setups around, but one thing that seems interesting is running two SMPs at once. Assigning 2 cores to each client and using some kind of affinity optimizer sounds intriguing. I know it'll eat a bit into the GPU WUs/day, but has anyone played around with this kind of setup?

[Elysium]

Sossaman

Quad core, and VERY little heat/energy.

[PTRMAN]

Thanks, but I was looking more for some recommendations for my specific system - I'm not running out to buy new CPUs or anything.

[Elysium]

My bad, sorry!

[polobunny]

SMP + GPU. Pretty much what you're doing currently.

[sbohdan]

Originally Posted by polobunny

SMP + GPU. Pretty much what you're doing currently.

I second that. you can OC your GPU and CPU and that would raise your PPD but make sure the OC is stable and the temps are OK otherwise you just ask for more errors...

[PTRMAN]

That's what I figured. The problem is that OCing would require me to actually stop folding for a while. I wonder what kind of PPD improvement I might expect to see - right now I'm running at 2.4; If I can reach 3.0, which seems to be fairly stable based on what I've read from other folks (YMMV of course), What kind of improvement would that be - should I expect roughly a 25% gain (2.4 to 3.0)? or something far less significant?

[Dread]

There really is no set number (percentage of improvement oc product X)
because there are too many variables still. OC'ing your quad to 3 GHz will show an improvement in returns, but we can't quantify it with a solid number of 25% etc. This is a direct result of the specific work units that are assigned, what the computer is doing (or not) doing while folding that wu, the network health (affects upload completed wu and download of new ones) and a host of other things.

Bottom line: If you can get a stable oc on it, it WILL increase your ppd.
By how much? Only you will be able to tell us that, and then only for individual work units.

[polobunny]

Originally Posted by PTRMAN

That's what I figured. The problem is that OCing would require me to actually stop folding for a while. I wonder what kind of PPD improvement I might expect to see - right now I'm running at 2.4; If I can reach 3.0, which seems to be fairly stable based on what I've read from other folks (YMMV of course), What kind of improvement would that be - should I expect roughly a 25% gain (2.4 to 3.0)? or something far less significant?

It's hard to say really.
In the end, it's always worth it though, even if your daily gain is only 10% PPD.
Let's say you do 5,000PPD. Then you take 1 full day to overclock your computer and fine tune everything.
Then the days after, overclocked, you do 5,500PPD instead of 5,000PPD. It will only take 10 days for you to get back to the score you should have been getting.
Between you and me it will probably be more than 10% gain too, but that was just to make you realize that even if you lose a full day fiddling with your overclock, it's likely worth it.

[PTRMAN]

Originally Posted by polobunny

It's hard to say really.
In the end, it's always worth it though, even if your daily gain is only 10% PPD.
Let's say you do 5,000PPD. Then you take 1 full day to overclock your computer and fine tune everything.
Then the days after, overclocked, you do 5,500PPD instead of 5,000PPD. It will only take 10 days for you to get back to the score you should have been getting.
Between you and me it will probably be more than 10% gain too, but that was just to make you realize that even if you lose a full day fiddling with your overclock, it's likely worth it.

I agree with your logic, but your assessment of my ability to get an OC to work in one day is perhaps optimistic. I've never done it before, and I'd heard these things are iterative, requiring lots of little adjustments until failure, followed by multiple hours of stress testing to ensure stability. If that's how itis done it'll take longer than a day. I still want to explore OCing, but now you see my dilemma.