Folding Forum

I used to fold -bigadv on my 4.5ghz 2600k i7 but now they dont work im just using -smp.
Im only getting around 7k ppd whats wrong with that?
Im using the v6.34 windows client and getting a tpf of around 8:30m to 9m

calculate it out. As to why, you need to do the work on the task manager to see if there is something else running. Folding is highly sensitive to other tasks taking CPU time and it is not even close to linear.

7k PPD for a 4.5GHz i7 2600K is very low. Is that just for a particular WU?

Is there a way of seeing what projects have been submitted without looking at the client log? Is it on some stats page somewhere if not i will have to wait till i get home.

Mind you my first two wu did have some downtime due to a bit of system instability, thats sorted now so wu number 3 should be much higher i hope, used to be getting about 44k a day with bigadv.

I think you need to see the log. How much of downtime? Points are calculated exponentially I believe, and if you have downtime, that can possible be the explanation for the low PPD.

Looks like it was due to the downtime.

Just picked up the 3rd WU a P6099 and its got a tpf of 7:07 HFM reporting points of 13997, that sounds a little better so PPD should be around 28k with that. Still not as good as the bigadv used to be.

Okay glad to know it's back to normal. Yes unfortunately bigadv increased the minimum cores to 16, it really is intended for very high end machines.

What did it matter if the WU were getting done fast enough?

Will my PPD generally drop if i do a gpu3 client as well, the GTX460 used to get me about 11k on its own.

Nvidia GPUs don't use much CPU time when folding. The SMP client will slow down a bit with the 460 running, but the PPD loss on the SMP side will be more than made up for by the PPD gain from the card.

robj20 wrote:What did it matter if the WU were getting done fast enough?

I guess they need faster returns.

robj20 wrote:Will my PPD generally drop if i do a gpu3 client as well, the GTX460 used to get me about 11k on its own.

It may drop but by a little amount only, the drop will be more than compensated by the gpu PPD. Nvidia GPUs don't use a lot of CPU cycles, while AMD GPUs use much more. So you can try folding with the GPU too. GPUs contribute much more processing power than CPUs so you'll do much more for the project

Well set a GPU3 off as well, only thing annoying about that is my graphics card has a horrible whine at 100%.

robj20 wrote:Well set a GPU3 off as well, only thing annoying about that is my graphics card has a horrible whine at 100%.

I am folding with 2 x GTX460
Both are currently working on project: 8008 WU's

#1 is OC to 800/1600/1600 and showing 77C with fan on auto at 88%
#2 is OC to 825/1600/1650 and showing 73C with fan on auto at 83%

I have never found the need to manually set the fans to run at 100%

I dont mean 100% fan, 100% load. Running at 820 core, 70% fan speed and 56c temp.
Its just the coils whining, can happen when not enough laquer is applied when manufactured.

robj20 wrote:I dont mean 100% fan, 100% load. Running at 820 core, 70% fan speed and 56c temp.
Its just the coils whining, can happen when not enough laquer is applied when manufactured.

Yeah tell me about it. My GTX260 sounds like a vacuum cleaner. Watercooling will solve the problem

iceman1992 wrote:

robj20 wrote:What did it matter if the WU were getting done fast enough?

I guess they need faster returns.

It has been explained in the FAQ and many times on the forum, but I'll say it again.

A very important segment of FAH's work is serial, not parallel. Project A (Run B, Clone C, Gen D+1) is created from Project A (Run B, Clone C, Gen D) and D+1 does not exist until D has been returned. The total time from D=0 to D=some large number matters and delaying any of the WUs delays the completion of the entire series. We're talking about years and years of SEQUENTIAL processing that must be completed either within the time a graduate student needs to complete his thesis or even within the lifetime of a senior researcher. You simply cannot predict the shape of a protein at some future time until you actually know it's shape at the present time so the faster you can process it, the quicker you'll get started on predicting the next change in shape.

Certainly things are learned from WUs with various values of A, B, and C, but a great deal of information depends directly on getting the values of D to a large enough value.

I'm not sure if FAH is unique in this regard or not. Certainly most distributed computing projects are purely parallel. The classic example is Seti@home, where the analysis of your part of the sky does not depend in any way on my completing the analysis of my part of the sky. Each WU is purely parallel and the time to complete a single WU doesn't matter, only that everything does get analysed.