Question OK< who is doing what in WCG ! ??

[Markfw]

So I have about 500 cores on WCG and my daily number is about 300k. The last 2 days, Zipspeed, biodoc and "the great cornholio" have been doing over a million a day ! what have I been missing ?

 

[Markfw]

The task lengths vary quite a bit. Did you get a bunch in the que or just the one?

Well, I have it running on 2 video cards right now, a 2080TI and a 3070, but each only appears to have the running task. Since BOINCTasks does not show these at all, its not easy to be sure.

 

[Icecold]

Well, I have it running on 2 video cards right now, a 2080TI and a 3070, but each only appears to have the running task. Since BOINCTasks does not show these at all, its not easy to be sure.

Is there any chance you have Boinctasks set to not show GPU tasks? If you right click the task list, there is a checkbox for 'Show GPU tasks'. Mine currently show up in Boinctasks.

 

[Endgame124]

I just looked at the WCG stats. Yesterday I did over a million. BOINCTasks is not showing these units for some reason, its been working apparently since yesterday.

My host did 1 gpu task at a time for a few hours in the evening. The next day I had dozens of tasks on the host.

 

[Markfw]

Is there any chance you have Boinctasks set to not show GPU tasks? If you right click the task list, there is a checkbox for 'Show GPU tasks'. Mine currently show up in Boinctasks.

Thanks for that ! Now it downloaded on 5 boxes, but the other 3 , even though it downloaded, were not running since the activity did not have it enabled. I turned it on those boxes so we would not waste the units. So now it running on a 1060, a 1070TI, a 2060, a 3070 and a 2080TI.

Edit: Nope now its on a 1080TI also, so 6 boxes. And about 200 total downloaded.

 

[Markfw]

The 3070 and the 2080TI are both running about 5 minutes a task (on the GPU tasks). The slowest of course is the 1060 at 25 minutes. The 1070TI is about 13 minutes, the 2060 is about 12.5, the 1080TI is about 10 minutes.

Tomorrow should be REAL interesting to see my WCG stats.....

 

[Icecold]

Most of my tasks on my 750ti are averaging about 14 minutes. I would expect a 750ti to be much slower than even a 1060 let alone the other cards. If I calculate number of tasks completed on prior days it seems to average out around 14 minutes per task as well. I wonder if I'm getting different tasks than you are.

 

[Markfw]

The task lengths vary quite a bit. Did you get a bunch in the que or just the one?

Its hard to count, but over 200 total task on 6 computers are queued up.

Edit: it also has 50 tasks queued up on a box that says DON'T use GPU

 

[StefanR5R]

it also has 50 tasks queued up on a box that says DON'T use GPU

Do you have a separate device profile set up and assigned to this computer?

Or is the GPU merely suspended locally on this computer? This alone won't prevent task fetching entirely.

 

[Markfw]

Do you have a separate device profile set up and assigned to this computer?

Or is the GPU merely suspended locally on this computer? This alone won't prevent task fetching entirely.

GPU is told nver to be used by BOIB, becuase its running F@H

 

[StefanR5R]

GPU is told never to be used by BOINC, because its running F@H

But how? There are several ways of doing it, and some of them don't prevent task fetching for GPUs.
If boincmgr -> Activity -> "Suspend GPU" is used, the client will generally still request work for the GPU(s).
If one or another of the related cc_config options is used, it too may still request work for the GPU(s), IIRC. Unfortunately I don't remember the details, because I rarely use these options.

The safest way to configure this is on the project server, in case of WCG as mentioned in the device profile. Which is a nuisance because different computers may need different profiles then.

 

[Markfw]

But how? There are several ways of doing it, and some of them don't prevent task fetching for GPUs.
If boincmgr -> Activity -> "Suspend GPU" is used, the client will generally still request work for the GPU(s).
If one or another of the related cc_config options is used, it too may still request work for the GPU(s), IIRC. Unfortunately I don't remember the details, because I rarely use these options.

The safest way to configure this is on the project server, in case of WCG as mentioned in the device profile. Which is a nuisance because different computers may need different profiles then.

BOINC manager, activity, suspend GPU.

Normally I only used GPU for F@H, and all the cpu's split between WCG and Rosetta. So now I have a mix, so I use this method.

 

[StefanR5R]

This explains it then. It's not completely intuitive, but boinc-client requests GPU work (if a project with GPU support is enabled) *even* if GPU computing is suspended (or/and if CPU computing is suspended).

The only requirement is that something causes the client to issue a scheduler request.¹ Can be a manual update, or can be a request which is associated with CPU work, e.g. CPU result reports.

Besides, likewise, the client requests CPU work (if a project is enabled) *even* if CPU computing is suspended. The only requirement is that something causes the client to issue a scheduler request.¹

¹) Only if the client determines that the work buffer is full, it does not request more work. Or if all projects are set to "no more work", of course.

You could try this in cc_config.xml:

<cc_config>
    <options>
        <!-- your other options here, if any -->

        <!-- no_gpus: If 1, don't use GPUs even if they're present. Requires a client restart. -->
        <no_gpus>1</no_gpus>
    </options>
</c_config>

Like the comment says, in order to apply it, restart the client after you added this to cc_config.xml. Likewise, if you remove it later or set it to 0, restart the client.

Note, I just took it from the documentation, I never tested it myself.

 

[Markfw]

I appear to be held back, only 2 computers are running GPU units. And a 3rd one has unit, but not processing them ????

 

[Markfw]

Did all the GPU units disappear ?

 

[StefanR5R]

There was a stress test of the WCG servers going on during which they issued a lot more GPU WUs per day than before. This test is now over, GPU WUs are supposed to be scarce again. (I haven't read up on the details. There is also a maintenance scheduled for tomorrow (or today where I am; starts less than 14 hours from now), maybe they have shut down GPU work generation early.