Only run long-runs task on that fast GPU. Deactivate short runs in your profile, but leave "if no work is available, accept work from other queues" checked so you won't run dry if at some point there are no long runs available.

Definitely reserve a CPU core, i.e. only run 3 CPU tasks in your case.

The current long runs are almost exclusively "GERARD_FX..". They result in relatively low GPU utilization on GPUs with many shaders, hence it's beneficial here to run 2 WUs concurrently. The best way to do this is via an app_config.xml. Do you know how to set this up?

Increase your GPU memory clock to at least stock clocks. Be default nVidia underclocks it for GP-GPU tasks.

You can overclock your GPU by setting a positive offset for the core clock in the usual OC utilities. This will lower GPU utilization, but it's faster. I have mine running at ~1.32 GHz and ~1.10 V. This is far below the maximum stock voltage of ~1.20, which makes the card work significantly more energy-efficient even with an overclock. I achieve this by reducing the power target. My card is working at 77% of 200 W (default power target for my model, this is probably different for yours), i.e. 156 W. GPU utilization is very high 92-93%, at least for Win Vista and newer.

And finally: you can dedicate an entire CPU core to feeding the GPU. This gains a few more % GPU utilization in my experience. You can do so by setting an environment variable "SWAN_SYNC" and giving it any name. Do you know how to do this?

I have 2 physical cores of my i7 3770K @ 4.0 GHz dedicated to feeding my GTX970, running 2 WUs concurrently. It needs a bit over 67k seconds for 2 long runs, i.e. would yield 655k RAC if it could run undisturbed.

MrS
____________
Scanning for our furry friends since Jan 2002

Hi robby,

to answer your PM:

1. The BOINC program is installed somewhere under C:\programs.. by default. The data is in a seperate directory in the user profile. To find it I suggest to search for the file "client_state.xml". This should be in the "root" of the BOINC data folder.

2. From there navigate to the folder "projects", then "www.gpugrid.net". That's we want to place the "app_config.xml".

3. Let's create a new .xml file here. A relatively easy way is to right-click some existing file, e.g. "slideshow_cellmd_00", then "open with" and choose "editor". Then choose "save as", change the file type to "all files" and type as name "app_config.xml". Done!

4. Add the contents of that file. Right click it, and choose "edit". Delete everything currently in the file. Then copy & paste these lines:

<app_config>
<app>
<name>acemdlong</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>acemdshort</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>

The 1st section is for long-runs, the 2nd one for short-runs. The 0.5 after <gpu_usage> signals to run 2 WUs concurrently. The 1 after <cpu_usage> signals BOINC to reserve 1 logical CPU core for each of those tasks. Set anything you see fit.. although 0.5 might be a good choice for your system. This way both tasks would in sum request 1 full core. BOINC would run 3 CPU tasks and there'd always be some spare time on the 4th core to feed the GPUs.

5. Save the filed modification.

6. Restart BOINC. Or in BOINC manager, advanced view, choose "options/reread config files"

7. Let us know how it went and what's the GPU utilization now :)

MrS
____________
Scanning for our furry friends since Jan 2002

Robby, if those instructions are not enough you'd need to ask more precisely where you encounter a problem. Otherwise I wouldn't know how to improve them.

MrS
____________
Scanning for our furry friends since Jan 2002

Good to hear it works. What'S the performance you're getting now, or the improvement?

Zoltan has just summarized on how to make the app use a full core over there. If you want to maximize GPU production this can be a worthy trade-off.

MrS
____________
Scanning for our furry friends since Jan 2002