Posts by Ghost Rider 51

1) Message boards : News : BOINC roadmap / wish list (Message 97158)
Posted 28 Mar 2020 by Ghost Rider 51
Post:
I started using Boinc with a single core CPU. Switching projects was the only way to participate in multiple projects. There was a drawback. Sometimes a task was almost finished to report it in time before the deadline and then projects were switched :-(
Now I have a multi core CPU and for me it makes switching projects unnecessary.
So my thing for the wish list is: don' t switch projects but select for each project how many cores (threads) it can use. If this can be made possible, I will participate in more than 1 project again.

There is also another thing that can be improved. The manager shows for each task how much time it will take to finish (estimated). I let Boinc use 30% of my CPU and those estimated times should be multiplied by 100/30 in my case.

There already is the ability to select how many cores/threads each project can use.
The app_config.xml file you create in the projects home directory can do that. Look at the manual here https://boinc.berkeley.edu/wiki/Client_configuration to see how.
2) Message boards : Questions and problems : GPU idling just because I set from 10 days to 1 day of work storage (Message 93505)
Posted 3 Nov 2019 by Ghost Rider 51
Post:
perhaps you might like to mention the ones you selected rather than the ones you didn't

What? This has nothing to do with other ones that are not selected.
The point is that my GPU is idled and it's not getting any work for that GPU. Not even Prime Grid which is the only project I want is getting any work "that Prime works with my second 980 GPU before".
Here's the kicker... I just set BOINC to 10 days of work and it grabbed more Prime to work with the 980 and seemed to gab other work as well. Since I have "Send work from any subproject if selected projects have no work" check to yes just to see that it's for the idle GPU sprong back to life. Note: I had 10 days of work and store up to an additional 10 days of work for years with all GPUs running just fine until I set it to 1 day of work "meaning it still had lots of work for 2 GPUS and BOINC just stopped the next day just for 1 GPU.

I don't understand what you mean by "ideal GPU".

I'm sorry I meant idle. My spelling is off.

This is for BOINC developers and I ran into what seems to be a bug that will idle a GPU.


I wonder if, because you have 2 GPUs that are different (980 and 1080), maybe one empties it's queue faster than the other (and the work units are different). Then, because the remaining work for the one still active is more than the 1 day limit you set, boinc may be delaying sending work until both GPUs are less than the set amount before it will send more work for the idle one.

I had a similar situation when crunching for 2 different projects in that boinc did not seem to ever balance the load between the projects. It would seesaw back and forth from doing work on one project 100% of the time to doing work on the other project 100% of the time and never was able to balance it (on one machine).
My solution was to dedicate one machine to one project and another to the other project.

It could be doing the same for the different GPUs. Maybe if those GPUs were in different hosts this conflict would not occur.

I agree that load balancing should be a developer priority.



Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.