Unaccountably, one of my systems is using one less CPU then it has assigned to it. I run WUProp and FreeHAL on all my systems. These are never assigned any cpu resources. I use the %75 processor assignment to limit all my quads cores to 3 cpus as I have found that gpu tasks get starved for data if all are assigned. I just noticed that ONE system was only using 2 cpus: project DNA was runnning only two work units. It should have been 3. I set %processor up to 100 and now DNA is running 3 cpus. It should be running all 4 if I understand how cpu assignments have been made.



The above shows that milkyway and primegrid use .97 and .89 percent of cpu each. That is over 1 full cpu. Is this a new feature of 7.0.18 that when it sees gpus using over 1 cpu that it limits cpu tasks? I had not noticed this before. But then, I never read the release history. If this is the case then I no longer need to set aside a cpu for the gpu any more.

Another system is running collaz on three nvidia boards. Collatz shows 0.01 cpu usage. That is even less than FreeHal or WUProp. That seems way too low. However, I assume it is correct and that might explain why all 3 cpus are being used in the following system (set to %75 processor, not 100%)



I plan on changing all my systems to use %100 processor (all cores) unless someone can point out a GPU project that will be starved for data.

Is this a new feature of 7.0.18 that when it sees gpus using over 1 cpu that it limits cpu tasks?

Yes. Since 7.0.7
David Anderson wrote:

The current policy is:
If a multicore app is running, don't overcommit the CPUs (i.e. don't schedule 4.1 threads on 4 cores).

This is because multicore apps may run inefficiently if the CPUs are even slightly overcommitted (at least, that was the case with AQUA).

We can reconsider this if there's evidence that the above assumption doesn't hold in general.

-- David

---