I have no idea what's behind your inconsistent runtimes; but even when working properly the current GW GPU app produces much lower credit numbers than the Fermi GPU app. Hopefully this situation will improve as new versions of the app are developed; in the mean time you can opt out of it by going into your project preferences and setting 'Run test applications?:" to No.
This just got complicated:
That "test" app gets me OpenCL-intel-gpu-beta which takes 11 minutes on my intel cpu. That is ok even with only double digit credits. However, if that gpu (the HD 4600 built into my CPU) ever gets a real task such as gamma ray binary search then it takes just under 4 hours which I don't want as it eventually times out. I need to restrict those GRB work units to non-intel gpu. This can be done with cc_config and/or app_config but it is a PITA as I got to figure out how to configure that file. There is a BOINC 3rd party GUI cc_config program but it does not do much
BOINC Client Configuration (2013/10/20 coproc & exclude_gpu not supported yet)
The problem is the apps are different for each project and having to look up the exact app name and the somewhat easier "OpenCL number" of the GPU is the real PITA. Anyway, it will get done, as I don't like idle time.
S9000 is an AMD GCN 1.0 (AMD Graphics Core Next 1) GPU architecture and so do suffer from the same bugs in the current GW GPU APP as all other older GCN GPUs (like Radeon HD 7xxx series, R9 260-280 series) which cause a HUGE slowdowns. About 10-30 folds from "normal" (expected from hardware specs) speed.
This can be done with cc_config and/or app_config but it is a PITA as I got to figure out how to configure that file.
If you updated to a more recent BOINC version, >> 7.0.2 you can use the gpu_exclude in the cc_config.xml file. The instructions are in the client configuration document.
The numbering of the gpu is simply by the numbering of the card in the Event Log at startup. For getting the application name you look in the client_state.xml file for the <app> section and then the <name></name>of the application to use for the application you want to exclude the gpu from.
DanNeely wrote:I have no idea
)
This just got complicated:
That "test" app gets me OpenCL-intel-gpu-beta which takes 11 minutes on my intel cpu. That is ok even with only double digit credits. However, if that gpu (the HD 4600 built into my CPU) ever gets a real task such as gamma ray binary search then it takes just under 4 hours which I don't want as it eventually times out. I need to restrict those GRB work units to non-intel gpu. This can be done with cc_config and/or app_config but it is a PITA as I got to figure out how to configure that file. There is a BOINC 3rd party GUI cc_config program but it does not do much
BOINC Client Configuration (2013/10/20 coproc & exclude_gpu not supported yet)The problem is the apps are different for each project and having to look up the exact app name and the somewhat easier "OpenCL number" of the GPU is the real PITA. Anyway, it will get done, as I don't like idle time.
S9000 is an AMD GCN 1.0 (AMD
)
S9000 is an AMD GCN 1.0 (AMD Graphics Core Next 1) GPU architecture and so do suffer from the same bugs in the current GW GPU APP as all other older GCN GPUs (like Radeon HD 7xxx series, R9 260-280 series) which cause a HUGE slowdowns. About 10-30 folds from "normal" (expected from hardware specs) speed.
Quote:This can be done with
)
If you updated to a more recent BOINC version, >> 7.0.2 you can use the gpu_exclude in the cc_config.xml file. The instructions are in the client configuration document.
https://boinc.berkeley.edu/wiki/Client_configuration#Application_configuration
The numbering of the gpu is simply by the numbering of the card in the Event Log at startup. For getting the application name you look in the client_state.xml file for the <app> section and then the <name></name>of the application to use for the application you want to exclude the gpu from.