Is it possible to restrict the number of cores to 25% when a APU/GPU WU is crunching, but allow 50% of cores to execute when non-GPU WU is / are being crunched?

Much asked for, but no, that isn't possible unless you work with two separate scripts or batch files and two global_prefs_override.xml files.

---

Bummers.

I'll just have to manually kludge it by manipulating preferences based on what's being crunched at the time.

As Jord said you can't restrict it to 25% or 50% but you can tell BOINC to reserve cores when running particular science apps via an app_info or app_config file. The app_config is easier. Below is a example one I use for Einstein OpenCL apps running on an Intel iGPU.

<app_config>
    <app>
        <name>einsteinbinary_BRP4G</name>
        <gpu_versions>
            <gpu_usage>1</gpu_usage>
            <cpu_usage>1</cpu_usage>
        </gpu_versions>
    </app>
</app_config>

This says that for the Einstein BRP4G app I want to allocate 1 GPU and 1 CPU to each GPU task. This overrides the project supplied values. If you wanted to reserve say 2 cores to each one then you would change the "cpu_usage" value to 2. Use Notepad or some other text editor (not a word processor) and save it as app_config.xml. It goes in the Einstein project directory under the BOINC data directory, so you can have separate ones for each project. You'll need to restart the BOINC client for it to pick it up.

Full details are on the BOINC web site Client_configuration under the section Application configuration

---

MarkJ

Hmmmm. Thanx for the reply. Seems as if this would need to be declared across the universe of projects one is participating in in. The prollem arose with the current batch of Einstein WU's I'm crunching, i.e., Binary Radar Pulser Search (Arecibo GPU) 1.39 (BRP4G-openCL-ati), which states: 0.5 CPU + 1 AMD/ATI GPU. What's happening is the CPU / APU core gets pretty warm and the fan is constantly spinning up / down.

I've changed the host-based preferences to restrict CPU to 0.25 and this results in only one other non-Einstein WU being crunched while the aforementioned WU's get crunched.

I'm assuming that two CPU are utilized with the single APU / GPU as adjunct to process these Einstein WU. The load on either core is less than if a single WU is tasked to a single core.

All that's fine. I'm just concerned that when the Einstein WU crunching gets down-prioritized, then only a single WU for some non-Einstein project would crunch, when in fact I'd be happy with two non-Einstein project WU's being crunched concurrently on separate cores.

So this custom app_info / app_config solution would be a means to such end.