I want to know if I can process two WU at same time in mi GPU (R7 Spectre). Its a APU A10-7700K. I have assigned RAM's 2GB to GPU.
Thanks for read me and happy crunching!
Here at Einstein you can do it thru your website settings under your account, preferences, project and then scroll down until you see the actual units you want to run, for those that you want to run 2 at a time put this in the box 0.50 for those you only want to run one at a time leave it at 1.0. Be sure to save changes at the bottom of the page.
Thanks for your answer. It did not work. I had to create and modify "app_config.xml".
Actually the preference setting described in the answer you refer to does work and probably would give a better performance outcome than the settings you have in your app_config.xml file. I had a look at your list of tasks for that particular host and the only tasks you have on board were all sent back on 22 September. When you make a change to the GPU utilization factor on the website, it only gets propagated to your host when new work is sent. If you had triggered a work fetch after changing the GPU utilization factor to 0.5 and received a new GPU task, your BOINC client would have then known to run all GPU tasks two at a time without you having to construct app_config.xml.
There are two things you should look at in your app_config.xml file. Firstly there is currently no GPU app for FGRPB1 so it shouldn't be included. Secondly, for both BRP4G and BRP6, running on an AMD GPU, you are highly likely to suffer poor GPU performance if you don't stop BOINC from allocating CPU tasks to at least once CPU core. Your value of <cpu_usage> of 0.1 would only free up one CPU core if there were 10 concurrent GPU tasks. For running 2 concurrent GPU tasks, you need a <cpu_usage> of 0.5. This is exactly what you would have had if you had persisted with a GPU utilization factor of 0.5 on the website and downloaded new work.
Set the GPU executables to a higher priority then the CPU executables and you won't have to dedicate an entire CPU towards the GPUs. Mine are always set to Below Normal and Idle. They will just take what is needed. I run 6 GPU WUs along site 8 CPU WUs of another project and I get the same GPU utilization with the CPU WUs running as with them idle.
The website config offers no benefit over your own app_config, only a negative as you'll have to wait for the changes to take affect. app_config is immediate on WUs downloaded at any time.
Pakal wrote:Hi everyone!I
)
Here at Einstein you can do it thru your website settings under your account, preferences, project and then scroll down until you see the actual units you want to run, for those that you want to run 2 at a time put this in the box 0.50 for those you only want to run one at a time leave it at 1.0. Be sure to save changes at the bottom of the page.
Thanks for your answer. It
)
Thanks for your answer. It did not work. I had to create and modify "app_config.xml". I leave the XML's content if someone need it like template.
<app_config>
<app>
<name>einsteinbinary_BRP4G</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>hsgamma_FGRPB1</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP6</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
</app_config>
Pakal wrote:Thanks for your
)
Actually the preference setting described in the answer you refer to does work and probably would give a better performance outcome than the settings you have in your app_config.xml file. I had a look at your list of tasks for that particular host and the only tasks you have on board were all sent back on 22 September. When you make a change to the GPU utilization factor on the website, it only gets propagated to your host when new work is sent. If you had triggered a work fetch after changing the GPU utilization factor to 0.5 and received a new GPU task, your BOINC client would have then known to run all GPU tasks two at a time without you having to construct app_config.xml.
There are two things you should look at in your app_config.xml file. Firstly there is currently no GPU app for FGRPB1 so it shouldn't be included. Secondly, for both BRP4G and BRP6, running on an AMD GPU, you are highly likely to suffer poor GPU performance if you don't stop BOINC from allocating CPU tasks to at least once CPU core. Your value of <cpu_usage> of 0.1 would only free up one CPU core if there were 10 concurrent GPU tasks. For running 2 concurrent GPU tasks, you need a <cpu_usage> of 0.5. This is exactly what you would have had if you had persisted with a GPU utilization factor of 0.5 on the website and downloaded new work.
Cheers,
Gary.
Wow! That's new for me, Gary.
)
Wow! That's new for me, Gary. I'll see how it works on the next tasks.
Thanks a lot, Gary. Cheers!
Set the GPU executables to a
)
Set the GPU executables to a higher priority then the CPU executables and you won't have to dedicate an entire CPU towards the GPUs. Mine are always set to Below Normal and Idle. They will just take what is needed. I run 6 GPU WUs along site 8 CPU WUs of another project and I get the same GPU utilization with the CPU WUs running as with them idle.
The website config offers no benefit over your own app_config, only a negative as you'll have to wait for the changes to take affect. app_config is immediate on WUs downloaded at any time.