Ok, with all things RX 580 / BOINC related seemingly smooth at this point, and with much further forum reading. It seems that from an E@H perspective I could dispose of my app_config.xml and control the apps with my "home" location from there. I currently can't locate the post but I think one of ya'll wrote it... about the + / - of using app_config vs using project prefs. Not an option for me on MW but looks like I could revert to project prefs here. My thought on this is when something at project level change I don't have an app_config to remember to change on a couple machines. I'll try to search forums again for the + / - config thread. Any differing opinions?
Ok, with all things RX 580 / BOINC related seemingly smooth at this point, and with much further forum reading. It seems that from an E@H perspective I could dispose of my app_config.xml and control the apps with my "home" location from there. I currently can't locate the post but I think one of ya'll wrote it... about the + / - of using app_config vs using project prefs. Not an option for me on MW but looks like I could revert to project prefs here. My thought on this is when something at project level change I don't have an app_config to remember to change on a couple machines. I'll try to search forums again for the + / - config thread. Any differing opinions?
Assuming this turns out to be a direction I go, what do I have to do AFTER renaming the E@H app_config file to get project prefs to take over?
Thanx, Skip
Easy answer: in the Boinc Manager go to Options and then near the bottom is 'read config files' click it and sit back and wait for your next batch of tasks to start using the settings. Don't worry if you get an error message saying you have some 'unknown file types' or something like that it's simply because you aren't running them so NO the config file doesn't know about them.
I believe Skip is asking to get away from using an app_config.xml for Einstein that he has to keep track of on several hosts.
The easiest method is to use the Project Preferences app configuration to set the gpu_usage for each gpu app. Downside is you inherit the default cpu_usage that the project has defined in the task generation template and it is never a full cpu core, always some fraction less.
This can have an unwanted effect of telling the client that you have unallocated resources available to start more tasks up and overcommit the cpu leading to elevated cpu_times compared to the run_times for tasks.
If you want to use the Project preferences, all you have to do is either move the app_config.xml out of the Einstein project folder or simply rename the file to app_conf.xml.alt or something similar so the client won't recognize the file as an actionable app_config.
I renamed the E@H app_config.xml to app_config.bkp after setting the factors I wanted for this machine on "home" location profile (I also have "school" for boxes that aren't currently attempting GPU). I have first done the re-read config files from the manager.
Didn't see anything change, as it should've been.
Then for the GW app I changed the GPU factor from .5 to .33.
I see no change (still running 2) after re-read, restart boinc-client and restarting machne. The existing GPU WUs are still running 2 at a time.
I may copy back the app_config file, make the change to .33 and see if that shows any difference.
Any thoughts on why my GW O3 GPU tasks don't seem to take to running 3 concurrently?
Skip
EDIT: Re-instituted app_config.xml with the .333 factor, a re-read and E@H went to running 3 GPU tasks. Will check the next few WUs to see what the time to complete did.
EDIT2: Run times (/3) started getting reallly long some I"m back to using prefs from site at .5
Hard to say. You should have not read any app_config for the project. Went back to the BOINC docs and read this:
Quote:
If you remove app_config.xml, or one of its entries, you must reset the project in order to restore the proper values.
Should have remembered that. Sorry. So you need to reset the project to read the configurations set from the Project Preferences.
I believe the O3MDF tasks are very VRAM hungry. You might not have enough memory to run 3X. Also, I think that app utilizes a lot more of the gpu compared to the other apps.
Also note that the web site preference changes don't apply to the already downloaded tasks. Only to the next tasks that are downloaded after the change. Older tasks are handled the way that preferences were set when those tasks were downloaded. You'd have to abort those tasks if you want to immediately see how new actually preferences work.
App_config changes are applied also to already downloaded tasks when you re-read that file or restart the client.
Also note that the web site preference changes don't apply to the already downloaded tasks. Only to the next tasks that are downloaded after the change. Older tasks are handled the way that preferences were set when those tasks were downloaded. You'd have to abort those tasks if you want to immediately see how new actually preferences work.
App_config changes are applied also to already downloaded tasks when you re-read that file or restart the client.
This doesn't appear to 100% true.
Scenario I just witnessed (again):
I had 3 GW tasks running on the card with E@H 'Home' prefs set to 0.33. These happened to be the last WUs in the que.
I changed the website 'Home' prefs to 0.50.
A few minutes later BOINC downloads 2 new GW tasks that show as using 0.5 AMD. The already RUNNING set of 3 immediately drops one of the ones that says 0.33 AMD and continues on running 2x (but they are still labeled as 0.33).
Now there is a partially completed (about 15%) WU that says 0.33 but is actually going to finish up concurrent with one of the new ones labeled as 0.5.
It seems once the server determines future WUs will be 0.5 AND it sends a couple new ones the local boinc-client will immediately start running 2 concurrently no matter what the WU is labeled.
Ok, with all things RX 580 /
)
Ok, with all things RX 580 / BOINC related seemingly smooth at this point, and with much further forum reading. It seems that from an E@H perspective I could dispose of my app_config.xml and control the apps with my "home" location from there. I currently can't locate the post but I think one of ya'll wrote it... about the + / - of using app_config vs using project prefs. Not an option for me on MW but looks like I could revert to project prefs here. My thought on this is when something at project level change I don't have an app_config to remember to change on a couple machines. I'll try to search forums again for the + / - config thread. Any differing opinions?
Skip Da Shu wrote: Ok, with
)
Easy answer: in the Boinc Manager go to Options and then near the bottom is 'read config files' click it and sit back and wait for your next batch of tasks to start using the settings. Don't worry if you get an error message saying you have some 'unknown file types' or something like that it's simply because you aren't running them so NO the config file doesn't know about them.
I believe Skip is asking to
)
I believe Skip is asking to get away from using an app_config.xml for Einstein that he has to keep track of on several hosts.
The easiest method is to use the Project Preferences app configuration to set the gpu_usage for each gpu app. Downside is you inherit the default cpu_usage that the project has defined in the task generation template and it is never a full cpu core, always some fraction less.
This can have an unwanted effect of telling the client that you have unallocated resources available to start more tasks up and overcommit the cpu leading to elevated cpu_times compared to the run_times for tasks.
You can make the change here: Project Preferences - gpu allocation for app types - bottom of page
If you want to use the Project preferences, all you have to do is either move the app_config.xml out of the Einstein project folder or simply rename the file to app_conf.xml.alt or something similar so the client won't recognize the file as an actionable app_config.
Then restart the client.
I renamed the E@H
)
I renamed the E@H app_config.xml to app_config.bkp after setting the factors I wanted for this machine on "home" location profile (I also have "school" for boxes that aren't currently attempting GPU). I have first done the re-read config files from the manager.
Didn't see anything change, as it should've been.
Then for the GW app I changed the GPU factor from .5 to .33.
I see no change (still running 2) after re-read, restart boinc-client and restarting machne. The existing GPU WUs are still running 2 at a time.
I may copy back the app_config file, make the change to .33 and see if that shows any difference.
Any thoughts on why my GW O3 GPU tasks don't seem to take to running 3 concurrently?
Skip
EDIT: Re-instituted app_config.xml with the .333 factor, a re-read and E@H went to running 3 GPU tasks. Will check the next few WUs to see what the time to complete did.
EDIT2: Run times (/3) started getting reallly long some I"m back to using prefs from site at .5
.
Hard to say. You should have
)
Hard to say. You should have not read any app_config for the project. Went back to the BOINC docs and read this:
Should have remembered that. Sorry. So you need to reset the project to read the configurations set from the Project Preferences.
I believe the O3MDF tasks are very VRAM hungry. You might not have enough memory to run 3X. Also, I think that app utilizes a lot more of the gpu compared to the other apps.
Also note that the web site
)
Also note that the web site preference changes don't apply to the already downloaded tasks. Only to the next tasks that are downloaded after the change. Older tasks are handled the way that preferences were set when those tasks were downloaded. You'd have to abort those tasks if you want to immediately see how new actually preferences work.
App_config changes are applied also to already downloaded tasks when you re-read that file or restart the client.
Thanx all.
)
Thanx all.
Harri Liljeroos wrote: Also
)
This doesn't appear to 100% true.
Scenario I just witnessed (again):
I had 3 GW tasks running on the card with E@H 'Home' prefs set to 0.33. These happened to be the last WUs in the que.
I changed the website 'Home' prefs to 0.50.
A few minutes later BOINC downloads 2 new GW tasks that show as using 0.5 AMD. The already RUNNING set of 3 immediately drops one of the ones that says 0.33 AMD and continues on running 2x (but they are still labeled as 0.33).
Now there is a partially completed (about 15%) WU that says 0.33 but is actually going to finish up concurrent with one of the new ones labeled as 0.5.
It seems once the server determines future WUs will be 0.5 AND it sends a couple new ones the local boinc-client will immediately start running 2 concurrently no matter what the WU is labeled.
Skip
^ OK, good to know.
)
^
OK, good to know.