is there a way to run just 3 rosettas on a 4 core cpu as it starves the GPU units I run

is there a way to run just 3 rosettas on a 4 core cpu as it starves the GPU units I run

I have the following XML in the Rosetta directory. The following should limit minirosetta jobs to a maximum of 3.

<app_config>
<app> <name> minirosetta </name> <max_concurrent> 3 </max_concurrent> </app>
</app_config>


Here is the path of the directory on my Windows machine:
C:ProgramDataBOINCprojectsboinc.bakerlab.org_rosetta


Some project have multiple named tasks and you would need 1 line for each one. If you misspell the task name, the project will display the list of valid name in the MESSAGE log.

is there a way to run just 3 rosettas on a 4 core cpu as it starves the GPU units I run

I have the following XML in the Rosetta directory. The following should limit minirosetta jobs to a maximum of 3.

<app_config>
<app> <name> minirosetta </name> <max_concurrent> 3 </max_concurrent> </app>
</app_config>


Here is the path of the directory on my Windows machine:
C:ProgramDataBOINCprojectsboinc.bakerlab.org_rosetta


Some project have multiple named tasks and you would need 1 line for each one. If you misspell the task name, the project will display the list of valid name in the MESSAGE log.

Or you could just restrict your CPU usage to 75% on a logical 4 core processor.

is there a way to run just 3 rosettas on a 4 core cpu as it starves the GPU units I run

I have the following XML in the Rosetta directory. The following should limit minirosetta jobs to a maximum of 3.

<app_config>
<app> <name> minirosetta </name> <max_concurrent> 3 </max_concurrent> </app>
</app_config>


Here is the path of the directory on my Windows machine:
C:ProgramDataBOINCprojectsboinc.bakerlab.org_rosetta


Some project have multiple named tasks and you would need 1 line for each one. If you misspell the task name, the project will display the list of valid name in the MESSAGE log.

Or you could just restrict your CPU usage to 75% on a logical 4 core processor.

I like your way better. I was thinking that the GPU job might be allocated a fraction of a CPU and the BOINC scheduler would idle another CPU ... leaving only 2 Rosetta CPU jobs running. The schedule makes my brain hurt.

But isn't the xml method Rosetta only, whereas the BOINC setting will mean that if the GPU project is BOINC-based then it'll not use that core either...

---

But isn't the xml method Rosetta only, whereas the BOINC setting will mean that if the GPU project is BOINC-based then it'll not use that core either...

No, that setting is for the CPU projects only, but does not affect the GPU support. For example, I have my "use at most % of the CPUs" set at 50%, and it uses 4 (out of 8) CPU cores on the CPU projects (e.g. 2 on Rosetta and 2 on DENIS at the moment), while another core is free to support my GPU on POEM. So I have 5 cores total in use on BOINC projects.

---

But isn't the xml method Rosetta only, whereas the BOINC setting will mean that if the GPU project is BOINC-based then it'll not use that core either...

No, that setting is for the CPU projects only, but does not affect the GPU support. For example, I have my "use at most % of the CPUs" set at 50%, and it uses 4 (out of 8) CPU cores on the CPU projects (e.g. 2 on Rosetta and 2 on DENIS at the moment), while another core is free to support my GPU on POEM. So I have 5 cores total in use on BOINC projects.

The BOINC scheduling algorithm makes my head hurt.
I use it to control the GPU jobs too.
My xml script was for only Rosetta but a script could be put into each of the project directories to control workload.

I also use the xml scripts to run multiple Seti@Home jobs in parallel on my NVIDIA GeForce GTX 970 because GPU-Z indicates that my GPU was only 40% loaded with 1 Seti task. MilkyWay will use 100% of the GPU no matter what GPU you have. Runtimes can get pretty short.

My xml suggestion answers his specific question about Rosetta but did not raise the "unasked" questions about tweaking the scheduling for other projects.

I tend to use the xml scripts to throttle particular projects to allow for more than one to make progress because I am constantly meddling.

My Seti script .... I left the 4 lines in to show how I "think" it works ... I ran 3 jobs for awhile but the average elapsed time per job did not change much beyond the 2 concurrent job setting so I left it at 2.

<app_config>

<app> <name>setiathome_v7 </name> <max_concurrent> 1 </max_concurrent> <gpu_versions> <cpu_usage> 0.3 </cpu_usage> <gpu_usage> 1.0 </gpu_usage> </gpu_versions> </app>
<app> <name>setiathome_v7 </name> <max_concurrent> 4 </max_concurrent> <gpu_versions> <cpu_usage> 0.3 </cpu_usage> <gpu_usage> 0.25 </gpu_usage> </gpu_versions> </app>
<app> <name>setiathome_v7 </name> <max_concurrent> 3 </max_concurrent> <gpu_versions> <cpu_usage> 0.3 </cpu_usage> <gpu_usage> 0.33 </gpu_usage> </gpu_versions> </app>
<app> <name>setiathome_v7 </name> <max_concurrent> 2 </max_concurrent> <gpu_versions> <cpu_usage> 0.3 </cpu_usage> <gpu_usage> 0.5 </gpu_usage> </gpu_versions> </app>

</app_config

My xml suggestion answers his specific question about Rosetta but did not raise the "unasked" questions about tweaking the scheduling for other projects.

Yes, it is inconsistent at best. I use "app_config.xml" to require each of my GPUs to use a single CPU core when I am running BOINC projects on all 8 cores of my dedicated machines, to prevent GPU starvation. But it works differently on Cosmology, at least when running their new "boinc2docker" betas, than on ATLAS, for example.

I am still sorting it out. You pretty much have to try it out and see what works best for your particular combination of projects; there may be no setting that is correct for all cases.

---

thanks guys - I just found a validate error issure on one box I bet its from gpugrid sharing a cpu core . every GPU app is running like a rocket with its own core