Hello everyone! So here's my problem:

I have noticed that I get a little more than 1 day off full workload from the servers for my 2 GTX 690's (4 tasks at a time), but then get only 2 tasks (half the full potential) for the next day. I have set network preferences as recomended:

Computer is connected to the Internet about every 0 days
Leave blank or 0 if always connected.
BOINC will try to maintain at least this much work (max 10 days). 	
Additional work buffer	0.05 days 

How can I get 4 tasks at all times?

thanks!

Going by the descriptions I think you are using old Boinc software.
I suggest you upgrade Boinc to 7.0.28, and also set your preferences to 0.05/0.05 or slightly higher (0.10).

If you don't already have <use_all_gpus>1</use_all_gpus> set in a cc_config file, do so,
FAQ - Best configurations for GPUGRID

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Sorry, I should have told that info before. I am running the latest BOINC client (7.0.28) and also have use_all_gpus flag to 1 in config file. I actually have "5" GPUs: the equivalent of 4 GTX 680 (two GTX 690, non-SLI) and 1 640 GT. I have also set GPUGrid to use only the four 680's (the two 690) by setting

<exclude_gpu>
<url>http://www.gpugrid.net/</url>
[<device_num>1</device_num>]
</exclude_gpu>

in the config file. All goes well, GPUGrid never utilizes the GT 640 (device 1 in my system) and the 4 remaining GPUs are used. Nevertheless, it uses the two 690s for about a day, and then "rests" for another day, using only 2 GPU's (I mean, 1 GPU in 1 690 and 1 GPU in the other 690). When this happens, the client is "smart" enough that it uses 1 GPU in each 690 card. I never leaves 1 card totally iddle.

I never "use" the 690s (for display, like games): their only reason for existence is crunching GPUGrid workunits ;). I'm stumped :S

P.S. The two 690 cards are actually physically linked for SLI, but is not enabled in the NVIDIA control panel. I am positive that this is not the cause of the problem as the it persists in a periodic, repeatable and totally predictable fashion. If this was the cause, like some kind of software glitch in the NVIDIA control panel, it would be totally random, dont you think?

This is as if there is some kind of GPU protection from the server side, and sending only half the workload after 1 day, but have never heard of this before.

I don't know if this helps or not, but I also have my two 670's in a SLI bridge, but also crunching/folding etc with SLI disabled and I have had no issues.

Maybe try restarting the BOINC client or going to the advanced options and clicking read config file?

This is as if there is some kind of GPU protection from the server side, and sending only half the workload after 1 day, but have never heard of this before.

No such protection is in operation on the server side.
Why don't you remove the GT640?

I don't know if this helps or not, but I also have my two 670's in a SLI bridge, but also crunching/folding etc with SLI disabled and I have had no issues.

Maybe try restarting the BOINC client or going to the advanced options and clicking read config file?

This has happened for days and computer restarts. I started researchign in the BOINC forums as this might be tied to a client issue. Will get back here if I find something usefull...

This is as if there is some kind of GPU protection from the server side, and sending only half the workload after 1 day, but have never heard of this before.

No such protection is in operation on the server side.
Why don't you remove the GT640?

I use that as the primary display device and for SETI@home too ;)

Why don't you remove the GT640?

I use that as the primary display device and for SETI@home too ;)

A GTX 690 isn't lagging if it is used for crunching and as the primary display device at the same time (I know it from my experience).
You lose much more credit by the idle GPUs, than you gain by a GT640.

What Retvari said is true about using the same GPU for crunching and for display at the same time. I do this with my computer and even play games while crunching, although this does slow down WU completion times, but I haven't had any issues.

I would still disable the physical SLI link. What's the point in having it, if it's not used?
I had trouble in the distant past with 4GPUs. I had to do things like extend the desktop to all 4 cards and use an omni-port. Recent drivers might have improved this situation, for two cards but not for four/five GPU's.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

It's been a week since my fisrt post and I haven't found a solution. I decided then to disattach the seti project in favor of this one and thus having only one project to work on. Hopefully, this will permit my system crunch 24/7 at full workload without strange work-fetch BOINC behaviour.

I have noticed that since BOINC 7.x, the algorithms for fetching work had a major overhaull (see e.g. http://boincfaq.mundayweb.com/index.php?language=1&view=592 and http://boinc.berkeley.edu/wiki/Release_Notes#BOINC_7_incompatible_with_Domain_Controllers) and some users have reverted back to pre 7.x versions or even configuring their systems for working with 2 clients on the same host.

Shouldn't BOINC developers be noticing this? For many other users, though, this is not a problem as many are attached to only one project and/or not using GPU as coprocessor, but from my perspective, the previous fetch algorithms did not cause trouble for them either. I would be happy to hear from someone who actually knows about this and hear from him/her the reasoning behind the changes made to 7.x

BOINC's 7.x work fetch algorithm - NO LIKE

attaching to 2 GPU projects with one client is not recommended. Causes problems getting correct amount of work from both.

---

XtremeSystems.org - #1 Team in GPUGrid