How do I get rid of this error trying to get more POEM WUs to download?

[DrPop]

So I just installed this 980 in Cheetah, and hooked it up to POEM. Only other thing I'm crunching right now is Enigma on CPU. When I try to get more WUs, it gives me this error in the event log:

1/19/2016 8:55:27 PM | Poem@Home | Not requesting tasks: don't need (CPU: not highest priority project; NVIDIA GPU: job cache full)

Every project I'm connected to is set at 100%, so I'm not sure what I've got set wrong. It's working, but only downloads 1 NVidia GPU WU at a time - so there is a lag for it to upload and download the next one when it finishes, and I don't like it sitting idle. heh...
Thanks for any help!

[zombie67]

Version of BOINC? Also, are you sure you are seeing all tasks? Sometimes I accidentally hit the "show only active tasks" in the manager.

[DrPop]

Hi Z, thanks for asking. BOINC Ver. 7.6.22 (x64) and yeah - I just tried that toggling the active vs all tasks, there's only ever one POEM WU downloaded at a time. Not sure why it will only give me one WU - something about the NVidia job cache being full, I guess, but I have no idea how to change that.

[zombie67]

For your nVidia project, what is the project weight? Zero maybe?

[FourOh]

Try reverting to BOINC 7.4.42. I'm not a fan of the scheduling on the 7.6.x versions... I see that message a lot when I need/want more work! You might also detach from other GPU projects, just as an additional measure.

[DrPop]

OK, thanks guys. Well, I tried upping the resource level on POEM to 1000%, with everything else left at 100%, and now, drum roll . . . it gives me TWO WUs! Dang, I almost fell over from it's generosity! So at least it seems to have one extra downloaded in the wings now...lol

[Mumps]

What about the cache size settings? Don't forget they've been monkeying around with what is meant by "Min cache/Extra Days" so I've had "Common Sense" settings like 3/3 lead to not getting work because of what it thinks that means these days.

[DrPop]

I've never touched the minimum before - that is set at 0. The max is set to 1 day or work. What do you think I should try instead?

[MindCrime]

Never touched the how many days of work setting in BOINC? It makes bunkering a whole lot faster if the project doesn't have an in progress limit. I'm sure it can vary by project but this is what BOINC recently said in regards to resource weight:

1/21/2016 12:56:39 PM | Leiden Classical | Requesting new tasks for CPU and Miner ASIC and NVIDIA GPU
1/21/2016 12:56:43 PM | Leiden Classical | (won't finish in time) Computer on 97.0% of time, BOINC on 100.0% of that, this project gets 8.3% of that

In my experience the message you're getting is because BOINC doesn't think it needs GPU work. Either the scheduler is just kind of off, a project is misrepresenting it's requirements or you actually have work for it.

This is similar to the problem that FourOH has mentioned in that he has ATLAS/VLHC limited concurrent tasks via app_config, but BOINC won't consider that those waiting WUs are only waiting on 1 thread, so your rig might not try to get work from other projects with idle cores because you have a bunch of ATLAS still even though they can only run one at a time.

Try a reboot if you haven't, make sure your openCL is all guda, cause AFAIK this project no CUDA. WHen boinc starts check the event log near the top to make sure you have an openCL device (gpu) not just 1 line for the CUDA device even though they're the same device.

[nanoprobe]

I've never touched the minimum before - that is set at 0. The max is set to 1 day or work. What do you think I should try instead?

Any 0 cache setting in BOINC is problematic at best. Depending on how much CPU work you want to cache you could try settings other than 0. Experiment with maybe .25 or .50 and see what happens.