GPU issues?
24 May 2016 5:17:43 UTC
Topic 198615
(moderation:
I really have no idea what I did. Well, I went to put my main computer on a location, instead of the default.
Now, it is running only 1 gpu task at a time despite the utilization factor being .5.
Additionally, "While BOINC is running, fraction of time computing is allowed 99.99%
While is BOINC running, fraction of time GPU computing is allowed 0.00%"
Well, that isn't true, its running a single gpu task right now.
https://einsteinathome.org/host/6212572
GPU issues?
)
i noticed that boinc is ignoring the gpu task settings too
what i did to make it take the new config is i deleted the einstein project xml file and restarted boinc it then will pull the new configs and wipe the project out .
from there you just add the project again in boinc it will load the right config and then einstein will resend you the lost tasks
Make sure you have the right settings in einstein preferences and boinc preferences
There might be other ways to fix that but this is what worked for me on multiple machienes
RE: I really have no idea
)
The reason for using 'locations' (also known as 'venues') is to allow different settings for machines in different locations. When you enable locations, each one starts off with the default settings. Before you assign a computer to a new location, you should check and change any settings for that location which are not what you want them to be. It's good practice to 'switch view' so that all location settings are displayed side by side. This make it easier to compare values between locations to see that they are exactly what you want them to be. If you did change the factor for the new location, perhaps you changed it after you had assigned the computer to the new location. If so, you would have to wait for new work to be downloaded before the new setting would be recognised.
This is nothing to do with settings or locations. Whilst you can see this information for your own hosts via the details page, it's not visible to others. I know what you are talking about because it's exactly the same on a details page for one of my hosts. I guess it's a BOINC deficiency, maybe something to do with the version of (highly modified) server code that this project uses. It might be client related. You could try upgrading to the latest client.
If you are concerned about the loss of production by running GPU tasks singly, you might also want to know that the cuda32 version of the BRP6 app you are using is quite a bit slower than the cuda55 version. If you "allow test applications" in your settings, you will be sent the new version and you will see faster completion times.
Cheers,
Gary.
RE: i noticed that boinc is
)
If by "task settings" you are referring to the GPU Utilization factor, I can assure you it's not being 'ignored'. That setting only arrives at your machine when your machine makes a request for new GPU work. If you are impatient, you can trigger an immediate work fetch by making a small increase in your work cache settings (days of work) locally through your BOINC Manager and then immediately revert the change once a work fetch has occurred. If you're not talking about GPU utilization factor, what "gpu task settings" are being ignored?
Please don't suggest such a drastic (and unnecessarily wasteful) course of action. I can't imagine any preference setting change that would not be implemented without doing this. You waste your own internet bandwidth which you may not care about but you also put unnecessary extra load on already stressed servers.
If you can't get something to work, just ask and someone will respond with a much less drastic solution.
Cheers,
Gary.
Oh please what is this
)
Oh please what is this 1992?
I don't think that a couple megs will break the server it's 2016
What if the forums become a lot more used are you going to tell them that they cannot do that because it might tax an already strained server more?
if you can't keep up with a couple more megs of utilization you doing it wrong
I have "allow test
)
I have "allow test applications" selected, but I am only getting Cuda 32. This all started with me poking around the settings for that... oops!
It did start running 2 tasks again however. At one point I did add another day to the cache, but because it was running only 1 task, I guess that never prompted Boinc to ask for more work, it already had too much.
RE: I have "allow test
)
When something like this happens - you made a change that doesn't seem to be working - it's always a good idea to have a look at what the scheduler thought about the change. From your account page click the 'view computers' link and for the host in question, click the 'last contact' link in the far right hand column. You might think it a bit daunting, but try to read through the stuff until you find something about checking the various GPU apps. Here's the bit in question from your last contact that refers to the cuda55 app version.
The problem is that your driver doesn't support cuda55. You need to install a more recent driver version.
Cheers,
Gary.