The GPU can only be maxed out if you config to at least allow 2 HCC-GPU tasks to run on 1 GPGPU, simply because the tasks have discreet CPU and GPU phases in the order:

1) CPU
2) GPU
3) CPU
4)CPU
5) GPU
6) CPU

Read up on app_config.xml posts and the required client 7.0.40 and higher for easy utilization. The sample below assumes 0.5 which is 1 : 0.5 = 2 per GPU and 1 : 0.5 = 2 GPU tasks assigned to 1 CPU core. Max_concurrent is optional if you wish to limit the overall number of HCC1 that can run on a system [GPU + CPU versions of the science]

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>12</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

The wiki: http://boinc.berkeley.edu/wiki/Cc_config.xml