Hello!

Excuse me please if my English is not right.

If I understand right, distribution for CPU and GPU Time between all task managed by section client_state.xml (and server reply file), that shown below:

<app_version>
<app_name>milkyway</app_name>
<version_num>21</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.684651</avg_ncpus>
<max_ncpus>0.684651</max_ncpus>
<flops>310080000000.000000</flops>
<plan_class>ati13ati</plan_class>
<file_ref>
<file_name>milkyway_0.21_windows_intelx86__ati13ati.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>brook32_ati.dll</file_name>
<open_name>brook32.dll</open_name>
<copy_file/>
</file_ref>
<coproc>
<type>ATI</type>
<count>1.00000</count>
</coproc>
</app_version>

When I manually correct values that marked bold, to 0.5 for exclusive using of 1 (and only one) CPU core for 2 GPU tasks that runs simultaneously, on first several units I see an ideal "picture" when BOINC runs 2 units on GPU and percent of GPU load all time was 99-100%. When one of units was ended, GPU loaded "decreased" to 97-98%.

When GPU crunch only 1 WU its load down to 0% on 2 or 3 seconds (when BOINC finish one WU and start next).

But when BOINC start to crunch next WU's values of CPU% reset to 68% and "ideal picture" was destroyed.

Can project administrators set the default values of this parameters to 0.25 for 4-GPU units crunching on 1 CPU core?

Happy crunching! :-)