Check out the newest addition to the family....

SG Distributed Computing Teams discussion, S.E.T.I., Folding@Home, as well as other types of cloud/distributed computing, AI, machine learning, cryptocurrency mining.
Post Reply
User avatar
goobee
Advanced Member
Posts: 772
Joined: Sat Apr 13, 2002 12:55 am
Location: Sunny California

Check out the newest addition to the family....

Post by goobee »

:D

Image
"Signature?........We don't need no stinkin' Signature!!!"
User avatar
Humboldt
Posts: 28212
Joined: Wed Oct 04, 2000 12:00 am
Location: Northern CA

Post by Humboldt »

I'm jealous.

What's it going into?
User avatar
goobee
Advanced Member
Posts: 772
Joined: Sat Apr 13, 2002 12:55 am
Location: Sunny California

Post by goobee »

My 8-core Xeon e5430 box. I've been running 2 SMP clients and 1 GPU client on a 8800GTS on this box. Let's see what kind of boost I get with the GTX285. :nod:
"Signature?........We don't need no stinkin' Signature!!!"
User avatar
OSULLY
Senior Member
Posts: 1347
Joined: Thu Nov 08, 2007 11:21 pm
Location: Catskills NY

Post by OSULLY »

:thumb:
___________________________________________
User avatar
cajundon
Posts: 122
Joined: Wed Nov 09, 2005 1:58 am
Location: Eunice, LA Cajun Country

Post by cajundon »

Hi Goobee,

How is it going with your new GTX 285, and how much PPD are you getting with it?

I'm saving for my next purchase of a GTX 295 + a 1200 Watt PSU, which should boost me to over 60,000 PPD with my folding farm. If I can afford the huge electricity bill, with running the AC most of the year to keep them cool, as I have to do already in Feb., with temps in the upper 70s already.

I'm running 3 9800 GX2s+ 1 8600 card, for a total of 7 GPU clients, + 5 PS3, and 3 SMP clients.
:thumb: Folding for a better future!

See you later alligator. Cajun Don.
User avatar
goobee
Advanced Member
Posts: 772
Joined: Sat Apr 13, 2002 12:55 am
Location: Sunny California

Post by goobee »

Super disappointing at the moment. Fahmonitor shows it at 6218.37 PPD. WTF?? :mad:
"Signature?........We don't need no stinkin' Signature!!!"
User avatar
joecool169
Posts: 805
Joined: Fri Nov 16, 2001 10:52 pm
Location: Ohio

Post by joecool169 »

goobee wrote:Super disappointing at the moment. Fahmonitor shows it at 6218.37 PPD. WTF?? :mad:
I have found on my own boxes that the smp slows the gpu way down no matter what I try. Could that be the cause?
Joe
User avatar
goobee
Advanced Member
Posts: 772
Joined: Sat Apr 13, 2002 12:55 am
Location: Sunny California

Post by goobee »

I cranked up the overclock on the shader, it helped a little. Not a lot though.

Image

I will turn off one of the SMP clients and see what the results look like.
"Signature?........We don't need no stinkin' Signature!!!"
User avatar
joecool169
Posts: 805
Joined: Fri Nov 16, 2001 10:52 pm
Location: Ohio

Post by joecool169 »

I would be willing to bet that if you closed a smp client and made sure the gpu was configured to use all available cores your ppd would go way up on that gpu. This is based on my own experience.

Do you use NV_FAH_CPU_AFFINITY command?
Joe
User avatar
goobee
Advanced Member
Posts: 772
Joined: Sat Apr 13, 2002 12:55 am
Location: Sunny California

Post by goobee »

No, but I'm willing to give it a go. How do you use that command?
"Signature?........We don't need no stinkin' Signature!!!"
User avatar
joecool169
Posts: 805
Joined: Fri Nov 16, 2001 10:52 pm
Location: Ohio

Post by joecool169 »

From another forum where I got my information.
With core 1.07 or later, they introduced a new environment variable called NV_FAH_CPU_AFFINITY and it take a binary mask. A binary mask is just a programming trick to set the bits but for the majority, it may be a bit confusing so here is a full list to help you get the correct value :

Code:

0 = all cores
1 = core 0
2 = core 1
3 = core 0 + 1
4 = core 2
5 = core 0 + 2
6 = core 1 + 2
7 = core 0 + 1 + 2
8 = core 3
9 = core 0 + 3
10 = core 1 + 3
11 = core 0 + 1 + 3
12 = core 2 + 3
13 = core 0 + 2 + 3
14 = core 1 + 2 + 3
15 = core 0 + 1 + 2 + 3

Alternatively, there is also a simple way to determine the value :

Code:

Core 0 = 1
Core 1 = 2
Core 2 = 4
Core 3 = 8

If you want core 1 and 3 only, add 2 and 8 to get 10.
If you want core 0, 2 and 3, add 1, 4 and 8 to get 13.

For your information, if you have a quad-core, setting a value or 0 or 15 set the same affinity to all cores. If you have more than 4 cores, each additional core always take a 2^n value where n is the core number, starting from 0.

Now that you know how to determine the value to enter in the environment variable, we need to create that variable. Here is the simple steps (OS independent) :

-Go to start and run type in: sysdm.cpl then a system config panel should pop up.
-Under the Advanced tab click the environment variables button.
-Click new and under Variable name put NV_FAH_CPU_AFFINITY
-Under Variable value put the corresponding # you desire... Look above to determine the desired value.
-Walla, you have now setup the environmental variable hit Ok and close the window.
-Stop and restart the GPU2 client to make use of the new environment variable.

Your GPU should now utilize whatever cores you specified! I generally use 15 to allow it to use all cores. Then just check the box as indicated above in Xil's guide to not lock the GPU to a single core and your ready to roll.
I use 0 myself.
Joe
Post Reply