I know I've been posted a lot recently, please bear with me. I just noticed something weird. Was playing with ratio of cpu to gpus in the v0.41 I had set ave 1 cpu and max 1 cpu for each astropulse on the GPU (1 AP per GPU). Under tools the computer preferences, with multiprocessor use at most "50"% is what I had. I have a quad core and running 2 GTX 650 ti boost on this machine. Figuring to free up 2 cores for the 2 GPUs and using the other 2 cores for crunching. when I did this, ALL the cores stopped processing and the GPUs continued with [ ap (1 cpu+1 GPU)] When I changed the percentage use under tools back to 100%, 2 of the CPU cores started to process again and the GPUs continued as well. This makes me wonder, when we tell Bonic to limit the usage to say 2 of 4 cores, then does it only with those 2 cores for both crunching and assisting the GPUs? If so, then on my other computer were I have and 8 core and am only utilizing 5 of them, that means the other 3 are sitting there doing nothing? Because from this it looks as if Bonic is only using the percentage allowed to use for both CPU and GPU combined and not using the freed up cores for the GPUs. Make sense?

...it looks as if Bonic is only using the percentage allowed to use for both CPU and GPU combined and not using the freed up cores for the GPUs. Make sense?

Yes.

Your "ave 1 cpu and max 1 cpu for each astropulse on the GPU (1 AP per GPU)" makes BOINC run 1 less CPU application whenever a GPU application is running. But if you run out of GPU tasks, BOINC will run a CPU task instead. So on a quad core system that setting by itself would have ensured 4 tasks running.

The 'use at most "50"%' tells BOINC to ignore the other two CPUs. Had you combined it with ncpus settings of less than 0.5 for GPU tasks, an AP task on each GPU would have added up to less than 1 CPU needed to support the GPUs, and BOINC allows that partial CPU overcommit. Windows would of course have used the CPUs left over after BOINC started two CPU tasks.

Either method by itself ensures there's enough CPU availability to support GPU operations, both together are too much as you've seen.

                                                                  Joe

Well, I ended up leaving the ave cpu and max cpu at 1 and leaving my utilization at 100%. This way there is 1 cpu per gpu. I had moved them to avecpu 0.5 and maxcpu 1 but saw that the Astropulses were taking twice as long as they were before. Since I made the move, time as returned to what they were before. I'm back at work on that computer which is an 8 core, so now 3 cpus are assisting the 3 gpus and the other 5 cpus are crunching away. Since that computer does nothing but crunching I'm not worried about tying up a cpu for OS. Thanks for the info.