General question: What cases does "Running (2 CPUs)" cover when a usable GPU is discovered and assigned work?

Narrative:

I presume it means that the app has requested 2 CPUs or otherwise BOINC knows to assign multiple CPUs as available to the app, based on preferences. That makes the most sense and is usually the case when no GPU is present, except on Linux, where I see newly downloaded tasks occasionally start running, and previously running tasks relegated to Waiting to run. Note that I have set Switch Tasks to a very high value to prevent switching, and shut down the computers before that interval expires.

On my only Windows 7 box with a GPU, with only a single-project single task auto-assigned to 2 CPUs, (others suspended) the Remaining value does not decrement, and CPU utilization according to task manager shows as 1% or less. This is confirmed by low CPU temperature and lack of progress on the CPU task. But if I enable one of the suspended tasks, the 2 CPU task comes to life and "Remaining" decrements normally. The newly enabled task reverts to "Waiting...." after a short time and the original 2 CPU task then runs normally. This behavior can be recreated after multiple BOINC restarts including reboots. System is WIN7 SP1 on AMD Athlon 64 X2 5600+ 3 GB RAM. BOINC Manager 7.12.1 (x64) v 3.0.1

This seems like a bug but I could not find documentation of precise rules on how BOINC assigns CPUs. I have read elsewhere that some apps may not be well behaved so can't rule that out either. App is Milky Way at Home.

Your comments, including a known good app (SETI?) much appreciated.

Descriptions like "Running (2 CPUs)" are written by the local BOINC client running locally on your machine, in response to instructions present - either in the <app_version> sent by the server, in app_info.xml, or in app_config.xml. The last two files are not usually present, but are optional files you can write yourself.

I've never heard of a project requiring two CPUs for a single task running on a GPU. Some projects have multi-threaded CPU applications available - usually intended for 4-core or 8-core CPUs - but they are intended for CPU use only: any GPU present would be scheduled independently.

Without having more detail, and the names of the project(s?) involved, it's hard to generalise.