Guide to running multiple GPU work units

[KieX]

When you click "Read local prefs file" it reads from global_prefs_override.xml

So I think you'd have to close BOINC and re-launch for it to take effect

Ah. Gotcha.
(all out of thanks buttons)

 

[dhoshaw]

so now the question is can i get more Wus done by running 2 per gpu or 1 per gpu. right now its 5m10s for 1 Wu per card. I can feel a very expensive addiction coming into play here

My 480 can do 1 gpu wu in ~6 min and 2 (using .5 cpu) in ~8 min.

 

[t_ski]

My 480 can do 1 gpu wu in ~6 min and 2 (using .5 cpu) in ~8 min.

Using your numbers as an example for AthlonX2, most of us look at how often a WU is returned then. Before, you had one every 6 minutes, now you have one every 4 minutes. That's a good improvement.

For me, I have three 7970's running four WU's each, and they each take about 4 minutes each to complete. That means I have about 12 WU's returned in those four minutes, for an average of one every 20 seconds. If I change that to 6 WU's per card, that rate only goes down a little - somewhere about 18 seconds. Then you have to look at the balance of heat and noise for the return, and wonder if it's any improvement.

 

[Norton]

I have never tried it live. If you do I presume you'd need to just hit the "Read prefs" option from the menu. I rather stick to the safe method of closing to avoid errors though.

With live I was meaning w/o going through the whole reinstall route just to add or drop the # of gpu wu's, etc..

Would still shut down the BOINC manager to make the change and likely do a restart as well (just for S's and G's)

Thx again guys

 

[stinger608]

Okay, I have (had) the "big app file" from Xtremesystems here:

http://www.xtremesystems.org/forums/showthread.php?283509-Working-app_info-files

For some reason, and I have no frigging idea why, I would have to reboot 2 to 3 times daily in order to keep my rig and video card crunching! I would be fine for awhile and then it would just stop doing WU's.

Well I am tired of messing with this and just loaded up the simple App_info file from this thread. I hope that solves the issue.

If anyone has an idea of why it would stop working I am sure open to discussing this.

 

[johnspack]

Trying to figure this out too. I used the simple script and left it at .5 figuring I'd get 4 gpu wus total on my 2 cards. So far I've got 8 cpu wus and only 1 gpu wu, I must be doing something wrong.... I also notice the gpu one is going much faster. Should I increase cpu usage above 60%?

 

[KieX]

Trying to figure this out too. I used the simple script and left it at .5 figuring I'd get 4 gpu wus total on my 2 cards. So far I've got 8 cpu wus and only 1 gpu wu, I must be doing something wrong.... I also notice the gpu one is going much faster. Should I increase cpu usage above 60%?

As you're using mixed cards by default it uses the best one. To use both go to the BOINC data folder (normally: C:\Program Data\BOINC) and look for file "cc_config.xml". If it doesn't exist, create it but the contents should have the following:

<cc_config> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config>

As for why you're not getting multiple WU's, check the location of the app_info is inside the \Boinc\projects\www.worldcommunitygrid.org folder. If it is.. there may be something up with it, post it here so we can check. (Sometimes the formatting can change when you edit)

EDIT: As the multi-gpu question appears to have popped up quite a bit recently, I've included it on the OP guide

 

[johnspack]

Still learning.... added that new info to the cc_info file. Then found out I had to go to device control and check use gpu for projects above! Didn't see that the first time. Now have 4 gpu wus as well as 8 cpu wus going successfully now. Wonder how many wus my 2xx cards can handle....

 

[t_ski]

On and off I have my kids' rig running two WU's on their GTX 280.

 

[BUCK NASTY]

Ok boy's, got a issue. Running 3 wu's per card(6 total) on a quad core CPU. I have 1 WU that lagg's behind all the others. How do break up the "ncpus". I tried .6 and .7 but there was no noticeable increase for the lagging WU. What do ya think?

 

[mstenholm]

Ok boy's, got a issue. Running 3 wu's per card(6 total) on a quad core CPU. I have 1 WU that lagg's behind all the others. How do break up the "ncpus". I tried .6 and .7 but there was no noticeable increase for the lagging WU. What do ya think?

I have noticed the opposite maybe because I run less GPU units then full/partial CPU supported ones. There is often one that runs faster the the others (and get less points). The "run patten" seems to be that a number (or all) GPU units begin and end at the same time but as you noticed one is trailing or in my case one is leaping in front. PhD assignment...

 

[[Ion]]

Ok boy's, got a issue. Running 3 wu's per card(6 total) on a quad core CPU. I have 1 WU that lagg's behind all the others. How do break up the "ncpus". I tried .6 and .7 but there was no noticeable increase for the lagging WU. What do ya think?

I think that's probably what we have to deal with if we have more WUs running than cores. I'm running 12 GPU WUs on a quad-core i7 (although with HT enabled)

 

[BUCK NASTY]

Just added a 2nd WU to the HD6850 and looks like the cards production has increased by 60%

 

[[Ion]]

Just added a 2nd WU to the HD6850 and looks like the cards production has increased by 60%

60% is surprising. By going from 1 to 3 WUs on the HD7770s I get about a 50% boost or so. I've yet to see what I can manage with 4 WUs.

 

[HammerON]

I was wondering if you had increased your Wu per card to 4 yet...

 

[KieX]

After some questions by Buck Nasty and johnspack it clicked that the settings must have started from OP. Revised the values in the NVIDIA files so that the GPU / CPU is correctly set to the default value of 1 GPU WU per CPU thread. Updated!

Previous code was fine, just the values were not optimal. New settings:
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>

 

[BUCK NASTY]

After some questions by Buck Nasty and johnspack it clicked that the settings must have started from OP. Revised the values in the NVIDIA files so that the GPU / CPU is correctly set to the default value of 1 GPU WU per CPU thread. Updated!

Previous code was fine, just the values were not optimal. New settings:
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>

Thanks KieX! I've just been manually adjusting it up to now.

 

[HammerON]

Thanks KieX! I might have to see about adding WU's to the GXT 580's again. First go around didn't work to well. I wonder how many I should try to do on each card...

 

[BUCK NASTY]

Thanks KieX! I might have to see about adding WU's to the GXT 580's again. First go around didn't work to well. I wonder how many I should try to do on each card...

Got 3 going on the 570. You might be able to do 4...?

 

[[Ion]]

I was wondering if you had increased your Wu per card to 4 yet...

Set it up last night. However, the computer was off for a few hours today while I brought it home. I should know reasonably well within a few days how things are looking

 

[t_ski]

*cough* *cough* 6 *cough* *cough*

 

[HammerON]

t_ski - you think I should try 6....
Not a bad idea. I guess that would take-up all my threads on the 970

 

[[Ion]]

*cough* *cough* 6 *cough* *cough*

On a HD7770??

I really don't think that I should be running 18 GPU WUs on a quad-core CPU!

 

[t_ski]

Sorry - I thought you only had two GPU's in that rig

Here it is on the 3570K + 2 x 7770's

 

[HammerON]

Sorry - I thought you only had two GPU's in that rig

Here it is on the 3570K + 2 x 7770's

http://img.techpowerup.org/121211/6wu4u.jpg

Oh - you were responding to [Ion]. I thought you were commenting on my post