can someone tell me, is it ok for intel GPU WUs to take a slot from CPU slots?
When I start the intel GPU WU, one of the running CPU WU will be suspended.
Is there a way for me to use an extra slot for intel GPU WUs hence the CPU slots are not affected?
Thanks in advance.
This behavior is normal. The task needs 0.5 CPUs + 1 Intel GPU. This is why one of your cores cannot process CPU tasks anymore. There is no way to run the workunit only on the GPU, since some of the calculations need to be done on the CPU. If you dig a bit deeper in the forum you will find more detailed explanations for that.
Thanks.
Your reply is so quick; I'm editing my post when you replied.
I just want to kindly ask, is there a way to modify the CPU utilisation on GPU WUs? Currently I seldom use this computer and I want to make full use of it by running BOINC applying all the resources. I don't mind the side effect of full utilisation.
Yes it's is possible to control the amount of CPU that is scheduled to support the GPU app, but it's highly recommended to free at least one core on the CPU when running GPU apps. That's because the GPU is almost always faster than the CPU and one usually want to maximize the GPU utilization. For me the Intel GPU app will not fully utilize the Intel GPU if I load CPU apps on all cores, so for me it's more efficient to run with one core free. Use something like GPU-Z to monitor the GPU load while experimenting.
To take control of the app you have to create a file called app_config.xml in the project subfolder of the Boinc data folder. Look at the startup messages in the Boinc event log for the path to the data folder, then got to projects\einstein.phys.uwm.edu and create the file there. Use Notepad and make sure it's saved as .xml and not .xml.txt
In the file something like this should work:
[pre]
einsteinbinary_BRP4
1.0
0.5
[/pre]
controls the GPU usage and can be set to 0.5 to run 2 task at once.
controls the CPU usage and can be set to 1 to reserve a full core or something smaller to not reserve a core.
After the file is created open Boinc manager and click "Advanced -> Read config file" for the changes to take effect.
Remember to use GPU-Z to monitor GPU load and max that out first then start adding CPU tasks.
Just a little bit to add to Holmis great explanation: you have 4 physical cores and 4 provided via hyperthreading for a total of 8. If one of them was free, task manager would show 12.5% CPU being free (well, actually 12%). Sicne you're seeing a value of 10% you already have less than a full core free. So when you launch another CPU task the GPUs will suffer in performance.
Think about it.. what does "maximum utilization" mean? Giving everyone work or maximising the scientific output of your machine? If it's the latter then leave that core free, as the additional throughput provided by 1 more hyper threaded core is not much (much less than a full physical core), significantly lower than the throughput the GPUs would loose by doing this.
System: HD4000 @ 1.25 GHz, DDR3-2400 with good timings, running 2 concurrent Einstein WUs, GPU-Grid on the nVidia and some SIMAPs on the CPU for an overall load of 7 of 8 CPU cores.
Before: typical runtime 1168s with driver 9.something.. one of the last ones of this branch.
Now: typical runtime 1126s with driver 10.18.10.3379.
That's almost a 4% performance increase, which is nice if given for free. And should increase my real-world RAC by 340.
It has been several weeks since I upgraded my PC video from Intel integrated video to an ATI/Radeon R250 series graphics generator. Subsequent to the update, I have not received an ATI GPU task. My event log correctly reflects the updated system but there are two entries that make me wonder if some user interaction is required to receive GPU tasks. They are: "app version refers to missing GPU type intel_gpu", and " application uses missing GPU type intel_gpu".
Please advise what actions, if any, are required. If this issue should be reports elsewhere, please provide the addressee. Thank you. Peter
Check your Einstein@home preferences and make sure that "Use ATI GPU" is set to yes. The tasks that are sent to ATI GPUs are: Binary Radio Pulsar Search (Arecibo, GPU), Binary Radio Pulsar Search (Perseus Arm Survey) and Gamma-ray pulsar search #3 so while checking your prefs make sure to opt in for at least one of them. You can also run the new S6 CasA beta GPU app on an ATI GPU and for that you also need to opt in to run beta apps.
As to the messages of missing GPU type it's usually because there's still tasks for the Intel GPU in the cache or because you're using a app_config.xml file. If you've set one up then edit it to remove the reference to the Intel app, if you don't know anything about app_config.xml then your not using one and then the easiest way to get rid of the messages ought to be to set "No new tasks", finish all work, do a manual update to report all finished tasks and then reset the project. That should clear it.
Thanks for the prompt response. I had been under the impression that read config files and read local prefs file in the advanced tab would change my preferences but that was a bad assumption as I found out when I followed your recommendation. The preferences in Einstein@home still reflected the old configuration. Thanks for the help.
... The tasks that are sent to ATI GPUs are: Binary Radio Pulsar Search (Arecibo, GPU), ...
Note that tasks for this application are not continuously available; e.g. currently there aren't any. If you want to keep your GPU busy with Einstein@Home, select at least one of the other applications as well.
Thanks for the prompt response. I had been under the impression that read config files and read local prefs file in the advanced tab would change my preferences but that was a bad assumption as I found out when I followed your recommendation. The preferences in Einstein@home still reflected the old configuration. Thanks for the help.
The local prefs only override the computing prefs found online, the project specific prefs that I linked to can't be set locally and thus only gets downloaded when Boinc contacts the project scheduler.
The read config files option loads the local and optional config files, cc_config.xml and all app_config.xml if there are any files to read.
RE: RE: Hi, can someone
)
Thanks.
Your reply is so quick; I'm editing my post when you replied.
I just want to kindly ask, is there a way to modify the CPU utilisation on GPU WUs? Currently I seldom use this computer and I want to make full use of it by running BOINC applying all the resources. I don't mind the side effect of full utilisation.
Yes it's is possible to
)
Yes it's is possible to control the amount of CPU that is scheduled to support the GPU app, but it's highly recommended to free at least one core on the CPU when running GPU apps. That's because the GPU is almost always faster than the CPU and one usually want to maximize the GPU utilization. For me the Intel GPU app will not fully utilize the Intel GPU if I load CPU apps on all cores, so for me it's more efficient to run with one core free. Use something like GPU-Z to monitor the GPU load while experimenting.
To take control of the app you have to create a file called app_config.xml in the project subfolder of the Boinc data folder. Look at the startup messages in the Boinc event log for the path to the data folder, then got to projects\einstein.phys.uwm.edu and create the file there. Use Notepad and make sure it's saved as .xml and not .xml.txt
In the file something like this should work:
[pre]
einsteinbinary_BRP4
1.0
0.5
[/pre]
controls the GPU usage and can be set to 0.5 to run 2 task at once.
controls the CPU usage and can be set to 1 to reserve a full core or something smaller to not reserve a core.
After the file is created open Boinc manager and click "Advanced -> Read config file" for the changes to take effect.
Remember to use GPU-Z to monitor GPU load and max that out first then start adding CPU tasks.
Just a little bit to add to
)
Just a little bit to add to Holmis great explanation: you have 4 physical cores and 4 provided via hyperthreading for a total of 8. If one of them was free, task manager would show 12.5% CPU being free (well, actually 12%). Sicne you're seeing a value of 10% you already have less than a full core free. So when you launch another CPU task the GPUs will suffer in performance.
Think about it.. what does "maximum utilization" mean? Giving everyone work or maximising the scientific output of your machine? If it's the latter then leave that core free, as the additional throughput provided by 1 more hyper threaded core is not much (much less than a full physical core), significantly lower than the throughput the GPUs would loose by doing this.
MrS
Scanning for our furry friends since Jan 2002
At last I decided not to
)
At last I decided not to change the settings; it is reasonable to reserve some CPU resource for GPU WUs.
Many thanks to holmis and apes (?!) for the good explanations.
Performance increase due to
)
Performance increase due to driver update!
System: HD4000 @ 1.25 GHz, DDR3-2400 with good timings, running 2 concurrent Einstein WUs, GPU-Grid on the nVidia and some SIMAPs on the CPU for an overall load of 7 of 8 CPU cores.
Before: typical runtime 1168s with driver 9.something.. one of the last ones of this branch.
Now: typical runtime 1126s with driver 10.18.10.3379.
That's almost a 4% performance increase, which is nice if given for free. And should increase my real-world RAC by 340.
MrS
Scanning for our furry friends since Jan 2002
It has been several weeks
)
It has been several weeks since I upgraded my PC video from Intel integrated video to an ATI/Radeon R250 series graphics generator. Subsequent to the update, I have not received an ATI GPU task. My event log correctly reflects the updated system but there are two entries that make me wonder if some user interaction is required to receive GPU tasks. They are: "app version refers to missing GPU type intel_gpu", and " application uses missing GPU type intel_gpu".
Please advise what actions, if any, are required. If this issue should be reports elsewhere, please provide the addressee. Thank you. Peter
Check your Einstein@home
)
Check your Einstein@home preferences and make sure that "Use ATI GPU" is set to yes. The tasks that are sent to ATI GPUs are: Binary Radio Pulsar Search (Arecibo, GPU), Binary Radio Pulsar Search (Perseus Arm Survey) and Gamma-ray pulsar search #3 so while checking your prefs make sure to opt in for at least one of them. You can also run the new S6 CasA beta GPU app on an ATI GPU and for that you also need to opt in to run beta apps.
As to the messages of missing GPU type it's usually because there's still tasks for the Intel GPU in the cache or because you're using a app_config.xml file. If you've set one up then edit it to remove the reference to the Intel app, if you don't know anything about app_config.xml then your not using one and then the easiest way to get rid of the messages ought to be to set "No new tasks", finish all work, do a manual update to report all finished tasks and then reset the project. That should clear it.
Thanks for the prompt
)
Thanks for the prompt response. I had been under the impression that read config files and read local prefs file in the advanced tab would change my preferences but that was a bad assumption as I found out when I followed your recommendation. The preferences in Einstein@home still reflected the old configuration. Thanks for the help.
RE: ... The tasks that are
)
Note that tasks for this application are not continuously available; e.g. currently there aren't any. If you want to keep your GPU busy with Einstein@Home, select at least one of the other applications as well.
BM
BM
RE: Thanks for the prompt
)
The local prefs only override the computing prefs found online, the project specific prefs that I linked to can't be set locally and thus only gets downloaded when Boinc contacts the project scheduler.
The read config files option loads the local and optional config files, cc_config.xml and all app_config.xml if there are any files to read.