(can hear the yawns)

Seen it now so many times on my quad... Up to the 3rd checkpoint BOINCTasks from remote indicates near 99% efficiency, and then when that 3rd checkpoint is being compiled, written and the 4th job starts, CPU time just goes near halt for 10 minutes, the only thing incrementing the ''elapsed" time (think Jean diagnosed that very early after the phase kick-off). Within 10 minutes efficiency dropping to 93% or less for the whole task up until that point and than the slow crawl up begins again. Shorter job... 4-6 hours and never getting above an overall 94-95%. Do they run 8-10 hours or more, the longer, the higher the *per-task* end efficiency becomes. That's on 17k-18k results per day a whole pile of time being abyssed at the 3rd checkpoint... that's the one where ''credit'' can be build with the crunchers.

Somebody brought up multi-threaded again... just a second, low priority and maybe offload that effort onto a subsystem... could a few cores of the GPU do that... it's in the machine and not being used 99% of the time and given the open endedness of this phase, 2 years more at least as mentioned by cleanenery, maybe an avenue to investigate... another option on the device profile "allow GPU/XXX subsystem use" :D

It's Friday afternoon and it's raining, no it's piping down.