Noticing something strange about how BOINC is counting the cores available, it isn't consistent.

Have two different 8 core machines each with a GPU, one PC, one MAC, same thing happening on each.

With 8 one core CPU tasks running a fractional CPU+GPU task will be run. So BOINC is using more CPU cores than exist, call the total 8.5 CPU. Everyone has seen this for a long time and it appears normal and expected.

I noticed when a Test4Theroy task started, it uses 2 CPU cores, that only 5 other one core tasks were running along with a fractional CPU+GPU task. BOINC was no longer using more CPU cores than exist, call the total 7.5 CPU.

I have also noticed when a milkyway N body task runs, is uses 8 CPU cores, that the fractional CPU+GPU task stops. BOINC was no longer using more CPU cores than exist, exactly 8 CPU.

Something is inconsistent. Have no idea why when a task asks for more than one CPU core a different way of counting how many cores are available is used. Did different people write different sections on BOINC to a different understanding of the requirements? Is it some kind of floating point round off issue that is a bug?

In any case inconsistencies are bugs.

---

Already fixed (six days ago):

Changeset dddf586 in boinc-v2

client: remove code that avoids overcommitting CPUs if MT jobs present.


This can lead to starving the CPUs if there are both GPU and MT jobs.
The basic problem is that a host with GPUs will never have all its CPUs
available for MT jobs.
It should probably advertise fewer CPUs, or something.

It'll likely be in the Next Recommended Release a year or so from now. ;-P

(I don't know if DA has any intention of backporting these little fixes into the Boinc 7.0.x line)

Claggy

Already fixed (six days ago):

Changeset dddf586 in boinc-v2

client: remove code that avoids overcommitting CPUs if MT jobs present.


This can lead to starving the CPUs if there are both GPU and MT jobs.
The basic problem is that a host with GPUs will never have all its CPUs
available for MT jobs.
It should probably advertise fewer CPUs, or something.

It'll likely be in the Next Recommended Release a year or so from now. ;-P

(I don't know if DA has any intention of backporting these little fixes into the Boinc 7.0.x line)

Claggy

Sounds to me like he is making the bug, over commit, even worse! And the "or something" sounds like he knows it!

---