Poem@Home released gpu applications at the end of december. It's currently available for linux (nvidia & amd) and windows (amd only). I've been running it for about a month now on my 5850.
Disclaimers:
Some tips that I've found useful in crunching:
Boinc 7.0.14 works fine for me with minimum work buffer set to 1 day and max buffer set to 5 days.
The windows app isn't very efficient on using the gpu so I had to set an app info file to run four work units at a time. My app_info file:
To help increase my chances of getting work I added <report_results_immediately>1</report_results_immediately> to my cc_config file so the client connects to the server as soon as a workunit is completed. My cc_config file has extra settings because I normally have 2 different gpu projects running at once but in the last week I have both gpus running poem.
The best version of the amd drivers that have worked for me is 12.1. It cut my cpu usage by almost half. Before using 12.1 each wu would use a full core, but with 12.1 it averages 60% of a core.
I'm going to be running poem until I hit 10 million at which point I'll move on to another gpu project. I've been looking at moowrapper for my 4830 and albert
for my 5850. 2012 will be a gpu year for me.
Disclaimers:
They've been having growing problems with the amount of work being done by the gpus. There are times when work isn't available. They have been upgrading their servers to handle the load but so far it hasn't been able to keep up.
- There is no applications for nvidia on windows, they are still working on it.
- The windows app isn't very efficient so each wu basically needs a cpu core to feed it. The gpu might not also be fully utilized. When running my 5850 only I averaged 56% utilization.
- To run gpu workunits you need to run at least boinc 7.0.
Some tips that I've found useful in crunching:
Boinc 7.0.14 works fine for me with minimum work buffer set to 1 day and max buffer set to 5 days.
The windows app isn't very efficient on using the gpu so I had to set an app info file to run four work units at a time. My app_info file:
Code:
<app_info>
<app>
<name>poemcl</name>
<user_friendly_name>POEM++ OpenCL</user_friendly_name>
</app>
<file_info>
<name>poemcl_0.1_windows_intelx86__opencl_ati_100</name>
<executable/>
</file_info>
<app_version>
<app_name>poemcl</app_name>
<version_num>1</version_num>
<plan_class>opencl_ati_100</plan_class>
<avg_ncpus>0.25</avg_ncpus>
<max_ncpus>1</max_ncpus>
<flops>2.1e12</flops>
<coproc>
<type>ATI</type>
<count>0.25</count>
</coproc>
<cmdline></cmdline>
<file_ref>
<file_name>poemcl_0.1_windows_intelx86__opencl_ati_100</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>To help increase my chances of getting work I added <report_results_immediately>1</report_results_immediately> to my cc_config file so the client connects to the server as soon as a workunit is completed. My cc_config file has extra settings because I normally have 2 different gpu projects running at once but in the last week I have both gpus running poem.
Code:
<cc_config>
<options>
<allow_multiple_clients>1</allow_multiple_clients>
<use_all_gpus>1</use_all_gpus>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>The best version of the amd drivers that have worked for me is 12.1. It cut my cpu usage by almost half. Before using 12.1 each wu would use a full core, but with 12.1 it averages 60% of a core.
I'm going to be running poem until I hit 10 million at which point I'll move on to another gpu project. I've been looking at moowrapper for my 4830 and albert
for my 5850. 2012 will be a gpu year for me.
Last edited:
Facebook
Twitter