I have a similar problem with one rig that is running a PCI 3 card in a PCI 2 slot. If the screen saver comes on, I get an error saying that the gpu is no longer working or is missing.

I also get the problem if I log in remotely an minimize the screen.

Finally what I did was start boinc on that machine as follows

c:\program files\boinc\boinc.exe --allow_remote_gui_rpc

Then I went to the Programdata directory and found the gui_rpc_auth.cfg file. The string in there is the boinc password. You can change this to something else I think. I just copied it.

Then back on the remote host, I logged in from boinc manager with that pw and everything is working fine. I do have to run 2 BM's to see everything but that's fine.

I just had to set the power options on the target machine so that there was no screensaver and the monitor never powers down. That's ok since it's on a KVM anyway and switching back and forth on that doesn't seem to matter.

Thanks for all the input, guys. A lot of the suggestions were not relevant because the power management settings suggestions were the same as I was already using, however I have been testing them to see if any one in particular might make a noticeable difference.

I have now set both machines to use non-Aero and will check again later to see if I get any more stuck units.

@ pfm3136, although I have multi GPUs in each machine, I'm not using crossfire, so that one can be easlily ruled out.

@ twilyth, interesting observations about remote connections. I had to stop using remote desktop on the GPU machines for this very reason. That was in the time of single image units, but I never got a situation like this until the advent of dual image units so I'm almost sure that it's not the culprit here.

re gui_rpc_auth.cfg and monitoring, instead of using two BMs you might want to conside BoincTasks which lets you monitor multiple machines in one interface. It is highly configurable and can make life so much easier. http://www.efmer.eu/boinc/boinc_tasks/

Thanks again to you all. It's comforting to know that I'm not the only one, even though it's so annoying because I never had these issues before the WUs changed from single units.

This project ran out of WU for 2 or 3 days.

That chart shows the time based on avarage of works per day and those days without work made a little mess on it. In a few days it will be ok. But i dont know why it is showed as PAUSED.

It's a mess, because the double image switch got the double whammy effect from the few days no new work shortly after. The "trimmed mean" works on a shorter period, so now it's 124 days ;P and "Paused" is now "Normal" [the old intertube cache view] :D It takes 10 days to get back to a "clean" average again. At the moment HCC is overweight [in total runtime] http://bit.ly/WCGQLK , so it might get a downward share adjustment [measured in results of total daily processed for WCG].

Trimmed Mean on MS office works by specifying a percent of high/low you want to exclude from the average. If the object is to remove only the single highest and lowest on last 30 days, the percent factor to dismiss is 2/30th i.e. 0.067. Enter 0.134 and it will dismiss the two highest and lowest, et cetera et cetera. At any rate, not going to touch the algo's, with the exception that from an arbitrary point I've assumed all results to be double image [to get back to the total completed images as a fraction of the overall estimated images that were guessed to be there at the last word of "so much added to...". Anyway, whenever Viktors has word on new estimates [he's the man doing the project duration planning], we'll surely hear. Still think this project will go intermittent at some point. There's always new sets of crystal being generated somewhere on the globe that can ** make use of the HCC system.

** edit: and have been making use of the HCC gateway to the grids processing power.

Coolstream: Thanks for the info on boinctasks, but I don't like to use anything not boinc supported. Lots of nice utilities come and go. What happens is you become dependent upon them and then one day they don't work any more and no one wants to step and support them.