Power Folding...?

[imaheadcase]

I've yet in all the years now GPU client has been released got it to work in conjunction with SMP cpu client. By not work, i mean it cuts the other in half speed, so its faster to run GPU over CPU. Assuming its a fast enough GPU

Even if i assign it one core on quad core i have, it still cuts HARD into the GPU client folding.

 

[petrusbroder]

Is some more optimization needed ... ?

 

[wbynum]

AMD cards (even the latest ones) are piss poor folding performers. Not worth running folding on them. Some even say when running Nvidia cards it cause to much of a bigadv performance hit to make it worth it. I haven't read to much into the details.

To give you an idea of good vs bad bigadv work units here are what my two i7 boxes are currently doing:

Good bigadv work unit:

i7 930@3.6ghz w/4gb memory: ~29.5k ppd, ~35 min per fold

Bag bigadv work unit:

i7 930@3.8ghz w/4gb memory: ~19.0k ppd, ~47 min per fold

It takes 100 folds in total to complete a WU. So even though the second computer is taking significantly longer with the slow WU. It's almost not worth it consider an i7 produces around 15k ppd with regular A3 WU's that take anywhere from 3 - 7 min per fold. Of course you have to take the good with the bad. Hopefully my next bigadv WU will not be one of these crappy ones.

 

[theAnimal]

I've yet in all the years now GPU client has been released got it to work in conjunction with SMP cpu client. By not work, i mean it cuts the other in half speed, so its faster to run GPU over CPU. Assuming its a fast enough GPU

Even if i assign it one core on quad core i have, it still cuts HARD into the GPU client folding.

Setting the GPU to a slightly higher priority should be all that`s required for Nvidia, and for ATI there are environment variables to reduce CPU usage.

 

[SemperFi]

What are you running as far as cpu/gpu? How much of a slowdown were you getting?

On a seperate note, I'm curious as to how I can track what the SMP client is doing as far as progress. With the gpu client I can just mouse over the system tray and see what my status is, but there is no icon for the SMP client that I know of. Anyone have any pointers for this?

2.87Ghz Phenom Quad core. ATI Raedon HD 4670 is my video card. The gpu client would crash daily from what I read may have been caused by all 4 cores crunching the smp client. I have several machines running putting up points. When I had the gpu client running I was averaging 32000-34000 points a week. I shut down the gpu client and just ran the smp on this my main rig and my weekly points jumped to 40000 a week. I haven't fired up the gpu client since I found out.

 

[brownstone]

2.87Ghz Phenom Quad core. ATI Raedon HD 4670 is my video card. The gpu client would crash daily from what I read may have been caused by all 4 cores crunching the smp client. I have several machines running putting up points. When I had the gpu client running I was averaging 32000-34000 points a week. I shut down the gpu client and just ran the smp on this my main rig and my weekly points jumped to 40000 a week. I haven't fired up the gpu client since I found out.

Interesting. I can't even get my quad core machine to do smp's. It freezes after about 2% and then I just get a bunch of error type messages as it tries to download and start again. Ultimately it just fails and closes. I have this experience whether my gpu client is turned on or off. So I ended up with the opposite problem, I just run my gpu. :\

 

[GLeeM]

Is it not worth running the GPU client in conjunction with the smp client on my i7 machine?

There are so many different combinations of hardware; and different WUs from F @ H, that it is hard to say for certain.

There are environment variables for ATI that help alot.
The early completion bonus is not as high on some WUs.

You may have to experiment for yourself to see if you get more ppd when running GPU. And then decide if the extra points are worth the electricity.

Example: My GTX275 gets ~9,300 ppd but it used to slow the smp by 4-5K ppd. I decided that the electricity used is worth it for now and especially during the winter when I have to pay to heat the house anyway. The PS3 will get turned on again during the coldest months.

 

[VirtualLarry]

Interesting. I can't even get my quad core machine to do smp's. It freezes after about 2% and then I just get a bunch of error type messages as it tries to download and start again. Ultimately it just fails and closes. I have this experience whether my gpu client is turned on or off. So I ended up with the opposite problem, I just run my gpu. :\

First of all, did you configure different MachineIDs for the GPU and the SMP client? If you don't, they will conflict.

Second, are you overclocked at all? Prime95/LinX/OCCT stable?

 

[brownstone]

First of all, did you configure different MachineIDs for the GPU and the SMP client? If you don't, they will conflict.

Second, are you overclocked at all? Prime95/LinX/OCCT stable?

I did configure different machine ID's for each and no oc has been done. I also tried disabling antivirus.

 

[brownstone]

Ok, time for a follow up. After uninstalling and reinstalling the smp client 2 more times I remembered that while I did not oc at all, I did undervolt my cpu quite a bit. My 965be has been running smoothly at 1.2v and even passed 7+ hrs of prime95. Just to "be safe" I bumped up the voltage to 1.2125. I wondered if this could be the issue so I bumped up my voltage to 1.25 and wonder of wonders....I'm now running stable SMP's. So thanks VirtualLarry for getting me thinking about the issue.

 

[brownstone]

There are so many different combinations of hardware; and different WUs from F @ H, that it is hard to say for certain.

There are environment variables for ATI that help alot.

What are these environmental variables for ATI and how do I manipulate them? What purpose do they serve (what do they help with)?

 

[theAnimal]

What are these environmental variables for ATI and how do I manipulate them? What purpose do they serve (what do they help with)?

These are the ATI environment variables:

FLUSH_INTERVAL
BROOK_YIELD
CAL_PRE_FLUSH
CAL_NO_FLUSH

They are used to reduce CPU usage by the GPU client. It's up to you to decide if the reduction in CPU production is worth the ppd for the GPU (which for ATI is usually not the case).

See here for a further explanation:

http://en.fah-addict.net/articles/articles-1-3+gpu-environment-variables.php

 

[brownstone]

thanks theAnimal for pointing me in the right direction.