Folding Forum

I'm stuck between 2 choices for F@H: do SMP or do GPU. The problem is that both aren't efficient enough for me to consider them useful.

The SMP client is limited due to the fact that i only have 2 cores, a Phenom II x2 550 BE, where the WU's perform much better on 4+. I'm crunching a 2 million step Wu, proj. 6701, that'll take me ~40-50 hours to do. my deadline is in 4 days and i'm only up to 33%, with 25 min per timestep approx. Last time, i had a 500000 step WU, and after 20 hours of crunching, i get a measly 500 points. I even put in a passkey for bonuses... guess i didn't turn it in in time for it? still, way too long for what i'm getting.

The GPU client, being ATi, is in obvious and already known terms on it's inefficiency. I can do a WU per 7-8 hours on my 4850, but the Science app itself is not efficient at all, so there is alot of wasted power that could have been in use with other GPU DC projcts. Have been waiting for at least SOME improvement with the GPU app since last year. GPU3 seems promising, but it isn't here yet for ATi, and with OpenCL still a vague mile away (as it is necessary for PG to rewrite code into CL..), i don't see it likely that it will come out by the end of this year. Plus, the fact that this GPU client requires a CPU DEDICATED to it (and with the fact that i only have 2 cores of CPU), it doesn't seem efficient to run an inefficient GPU client that is inefficient on utilizing an ATI card, which is also making my CPU inefficient in that i can only use 1 of the cores for other DC projt's. (The independent variables did help to reduce CPU usage... but not to the point where I can use it for other DC apps)

For now, i need to choose between SMP or GPU. uni-processor clients aren't favorable in this scenario, although i can do them. the yield for them isn't comparable with the GPU or SMP clients.

Cartoonman wrote:The SMP client is limited due to the fact that i only have 2 cores, a Phenom II x2 550 BE, where the WU's perform much better on 4+. I'm crunching a 2 million step Wu, proj. 6701, that'll take me ~40-50 hours to do. my deadline is in 4 days and i'm only up to 33%, with 25 min per timestep approx. Last time, i had a 500000 step WU, and after 20 hours of crunching, i get a measly 500 points. I even put in a passkey for bonuses... guess i didn't turn it in in time for it? still, way too long for what i'm getting.

For p6701, you have to upload the result within 4.9 days from the download time to get any bonus. If you have 67% still to do at 25 min/% that's 1.2 days which when added to however long it has taken you to get to 33% can be compared to the 4.9 days.

There are other reasons why you may not get a bonus and we can help you with that, too. Your most recent SMP WU was (P6021 R0 C150 G307) which was added to the stats database on 2010-10-17 15:09:45 for 476 points of credit. It was completed in 1.94 days so not getting a bonus is related to your passkey, not the deadline. Have you completed the required 10 SMP WUs with that passkey and that UserName, and is your completion rate above 80%?

The GPU client, being ATi, is in obvious and already known terms on it's inefficiency. I can do a WU per 7-8 hours on my 4850, but the Science app itself is not efficient at all, so there is alot of wasted power that could have been in use with other GPU DC projcts. Have been waiting for at least SOME improvement with the GPU app since last year. GPU3 seems promising, but it isn't here yet for ATi, and with OpenCL still a vague mile away (as it is necessary for PG to rewrite code into CL..), i don't see it likely that it will come out by the end of this year. Plus, the fact that this GPU client requires a CPU DEDICATED to it (and with the fact that i only have 2 cores of CPU), it doesn't seem efficient to run an inefficient GPU client that is inefficient on utilizing an ATI card, which is also making my CPU inefficient in that i can only use 1 of the cores for other DC projt's. (The independent variables did help to reduce CPU usage... but not to the point where I can use it for other DC apps)

Don't give up. See item #5 here The official word is that the "vague mile away" is a lot less vague than it has been.

As far as the dedicated CPU to run it, that depends on how OpenCL is implemented by ATI. I have seen no information on that subject, but ATI did recognize the problem more than two years ago ago and provided Environment Variables that could be used with the Brook version that do help a lot. It's reasonable to assume that since the recognized the problem then, they'd follow through with similar or better drivers for OpenCL -- but that's an unsupported guess on my part.

Well when you have the bonus, the SMP will be around 2200 PPD on the 6701.
Should be around 2600 PPD on normal SMP WUs.

Thats about what my E6300@3.6 gets, while folding on a GT 240.
When folding with ATI (which I haven't done in awhile) if the enviromentals are set correctly
it works too. Just not the same points with GPU folding.

You just have to spend the time to tweak it.

Well, the only issue at this point for the SMP is the fact that i cannot submit some (like 6701) WU's fast enough for the Bonus Due date, since i don't have my machine dedicated, although i try as much as possible to allot as much as possible for the SMP client. I haven't submitted 10 completed WU's yet, so that would be why for the Bonuses, at this point, but the turn-around time will factor later on.


@bruce: @ 25/TS, it's about 41 hours crunching time required. I don't dedicate my PC to DC, and i avg time per day to be bout... 6-9 hours weekdays, more on weekends, but not 24, VERY rarely. I plan later on, to possibly build a machine just for DC, but that won't be for a very long time.

I am hoping for an improved GPU client quite soon. I would love to help F@H with my GPU, but right now, it isn't worth it in my part to continue in inefficiency when i could use full potential in other DC proj's. When the long anticipated client comes out, i won't think twice to try it out.

Cartoonman wrote:...uni-processor clients aren't favorable in this scenario, although i can do them. the yield for them isn't comparable with the GPU or SMP clients.

Point-wise, the Classic Client may not match the SMP2 or GPU2/3. However, there are some scientific calculations that can only be done on the Classic Client which cannot be done on the SMP2 or even on the GPU2/3 so your contribution to science is very valuable.

Cartoonman wrote:...I don't dedicate my PC to DC, and i avg time per day to be bout... 6-9 hours weekdays, more on weekends, but not 24, VERY rarely...

That kind of contribution works well with Classic Client. Since your system is dual core, you can either run 1 instance or 2 instances, depending on your choice.

Cartoonman wrote:...I don't dedicate my PC to DC, and i avg time per day to be bout... 6-9 hours weekdays, more on weekends, but not 24, VERY rarely...

PantherX wrote:That kind of contribution works well with Classic Client. Since your system is dual core, you can either run 1 instance or 2 instances, depending on your choice.

I want to strongly support PantherX's suggestion. For somebody who folds 24x7 on a multi-core machine, SMP is a good choice. For somebody who folds 6-9 hours per day it's probably the WRONG choice. I believe that you'll find 2 instances of the uniprocessor client on a dual-core machine is the best choice.

I don't even want to guess what will happen to v7 or OpenCL for your gpu or when it might happen. All that new software will have some growing pains when it goes from being tested on maybe a dozen or two different configurations to every possible combination of hardware and software that happens run FAH. It'll take a while and it's still not certain to be the right choice for you.

If there are hardware upgrades in your future everything might change, but I'm going to assume not. One possibility for a tiny upgrade would be to leave your computer running 24x7. If the screensaver is set to "turn off the monitor" both the monitor and the GPU can go into power-saving mode. Your CPU will still be drawing some power since it will still be folding, but the extra cost of that electricity is small compared to hardware upgrades, and during the winter, the power goes mostly to warming the room which (depending on your climate) offsets part of your home heating costs. By running 24-7, you will be able to run the SMP client rather than the two uniprocessor clients. Anyway, think about it.

Unless you are doing intense cpu things, let it run. Other then playing BF2/BFBC2 I let it run. (quad)
Using my E6300 machine the same. Right now on the quad Q8200@2.88, email, web, on line jigsaw,
SMP - 6701 4200PPD, GTS250 - 11167 5333PD. Wife is on the E6300@3.6 SMP 6701 2350PPD, GT240 11167 3700PPD.

15K + while using the machines. Over 20K when not using.

@RAH: the prob with that is i have a dual core. you have a quad core. no matter how fast my 2 cores are, even at 4 ghz each, 4 cores @ 2.5 ghz each beats my CPU by 2 Ghz, hypothetically speaking. Nowadays, Ghz's are meaningless in comparing CPU's. It's now about how many cores, and the architecture and such. more cores=more performance per thread.


I'll be using the unicore client. I'm @ 78% and it's due in less than 24 hours. no way this is going to continue, bonus or not.

I thought you can't just add them?

Well I won't comment on ATI since I have switched.
But my E6300@3.6 does just fine, while folding on a GT240.
SMP = 2.4K PPD GPU = 5 K PPD

Close to 8 K PPD. Not bad to me.