Your 256Mb 8600GT doesn't have enough memory for the BRP4cuda32 class plan, and doesn't have new enough drivers for the BRP4cuda32nv301 class plan, even if it did, you'll then find it doesn't have enough memory for that app eithier
I noticed that before, that's why I disabled GPU in the cc_config.xml, in order to be sure to receive only CPU tasks.
Quote:
Is your cache full of CPU work from other projects?
With the new scheduler, I don't know how to check that??? But!!! I think you pointed out what's wrong.
As mentionned before, some projects sent many WUs in one shot, some of them, more than twenty WUs (???). With the previous scheduler, even in a such situation (to much WUs), all projects still updated themselves and asked (or not) for job alternatively and regularly. With the new scheduler, if ONE project saturates the global work reserve, other projects aren't even updated or refreshed, like suspended... To verify that, I suspended all projects except Einstein, and HOP! It sent me instantaneously 4 CPU tasks!
I did another test: as I mentionned before, when I changed for 7.0.27, I resetted none active projects. With the old scheduler, reset occurs progressively for all projects, within 15 mins, it was done. With the new scheduler, because too many work was already engaged, other projects haven't even start their reset... When I enabled again other projects, they started their reset. It's has been for more than a week that I updated for 7.0.27. A far too long time (from my point of view) to activate orders like "reset project" (which is independent of the amount of work a project has or hasn't)... So, it is well like the new scheduler suspend other projects...
So now, I really don't get it with this new scheduler??? How to set it up so that ~20 projects on a 4 cores CPU + GPU receive correctly and fairly ONE WU AT A TIME and fulfill the global work reserve? As it was done before pretty much efficiently...
RE: Your 256Mb 8600GT
)
I noticed that before, that's why I disabled GPU in the cc_config.xml, in order to be sure to receive only CPU tasks.
With the new scheduler, I don't know how to check that??? But!!! I think you pointed out what's wrong.
As mentionned before, some projects sent many WUs in one shot, some of them, more than twenty WUs (???). With the previous scheduler, even in a such situation (to much WUs), all projects still updated themselves and asked (or not) for job alternatively and regularly. With the new scheduler, if ONE project saturates the global work reserve, other projects aren't even updated or refreshed, like suspended... To verify that, I suspended all projects except Einstein, and HOP! It sent me instantaneously 4 CPU tasks!
I did another test: as I mentionned before, when I changed for 7.0.27, I resetted none active projects. With the old scheduler, reset occurs progressively for all projects, within 15 mins, it was done. With the new scheduler, because too many work was already engaged, other projects haven't even start their reset... When I enabled again other projects, they started their reset. It's has been for more than a week that I updated for 7.0.27. A far too long time (from my point of view) to activate orders like "reset project" (which is independent of the amount of work a project has or hasn't)... So, it is well like the new scheduler suspend other projects...
So now, I really don't get it with this new scheduler??? How to set it up so that ~20 projects on a 4 cores CPU + GPU receive correctly and fairly ONE WU AT A TIME and fulfill the global work reserve? As it was done before pretty much efficiently...
Cheers
I guess it might be best to
)
I guess it might be best to carry this discussion to the BOINC forum then, because it's more likely to receive attention by BOINC developers there.
Cheers
HB