Wow, I didn't expect all these replies. Seems I sparked a pretty good debate!
The GPU fails on stock settings and using folding clock settings [low core/mem, higher shader clock]. The cards are not unstable at these settings. Prior to changing OS I ran both cards full time for weeks with no errors using those settings.
If the GPUs failed during the WU run I would be inclined to agree. However I can restart my machine, pick up a "bad" WU and it will fail. Moreover, the WU doesn't DO anything. The GPU temps as monitored by GPU-Z barely even twitches from its idle temps. Regardless of temps the WUs will fail.
SBS 2008 is the only major change. The only hardware change is a PSU, which is a significant upgrade considering the last one, which as you know blew up. There was a brief thought that the PSU blowing did some minor damage somewhere, but I had errors before changing PSU, so I have ruled that out.
I am reluctant to believe that errors of this kind are down to me to solve on my own, especially when it is only certain Work Units. Only Stanford knows the ins and outs of their clients and Work Units. I cannot be expected to troubleshoot an error, when I am given absolutely no meaningful information. Presumably Stanford know exactly what went wrong but that data isn't reaching the folding forum mods or anyone else.
Perhaps there is a conflict between SBS 2008 and the CUDA drivers, that only manifests itself with certain WUs. I am told the problem projects are highly optimised, perhaps different parts of CUDA are being used which triggers the issue. Who knows? Maybe it is a bug/feature in SBS 2008, since it was not really designed for running high-end graphics. Again, who knows? Stanford certainly isn't saying.
The GPU fails on stock settings and using folding clock settings [low core/mem, higher shader clock]. The cards are not unstable at these settings. Prior to changing OS I ran both cards full time for weeks with no errors using those settings.
If the GPUs failed during the WU run I would be inclined to agree. However I can restart my machine, pick up a "bad" WU and it will fail. Moreover, the WU doesn't DO anything. The GPU temps as monitored by GPU-Z barely even twitches from its idle temps. Regardless of temps the WUs will fail.
SBS 2008 is the only major change. The only hardware change is a PSU, which is a significant upgrade considering the last one, which as you know blew up. There was a brief thought that the PSU blowing did some minor damage somewhere, but I had errors before changing PSU, so I have ruled that out.
I am reluctant to believe that errors of this kind are down to me to solve on my own, especially when it is only certain Work Units. Only Stanford knows the ins and outs of their clients and Work Units. I cannot be expected to troubleshoot an error, when I am given absolutely no meaningful information. Presumably Stanford know exactly what went wrong but that data isn't reaching the folding forum mods or anyone else.
Perhaps there is a conflict between SBS 2008 and the CUDA drivers, that only manifests itself with certain WUs. I am told the problem projects are highly optimised, perhaps different parts of CUDA are being used which triggers the issue. Who knows? Maybe it is a bug/feature in SBS 2008, since it was not really designed for running high-end graphics. Again, who knows? Stanford certainly isn't saying.
deleting the queue, changing machine ID and dropping clocks to minimum didn't help. A restart, on the other hand, did seem to sort it out 
