Since no new tasks were downloaded on my 4 hosts since yesterday late afternoon, I looked up the BOINC event log of the PC with 2 GTX980Ti and found the following entries:

4/14/2016 6:01:43 AM | | cc_config.xml not found - using defaults
4/14/2016 6:01:43 AM | | Starting BOINC client version 7.6.22 for windows_x86_64
4/14/2016 6:01:43 AM | | log flags: file_xfer, sched_ops, task
4/14/2016 6:01:43 AM | | Libraries: libcurl/7.45.0 OpenSSL/1.0.2d zlib/1.2.8
4/14/2016 6:01:43 AM | | Data directory: Z:\
4/14/2016 6:01:43 AM | | Running under account Administrator
4/14/2016 6:01:44 AM | | GPU detection failed. error code 1
4/14/2016 6:01:44 AM | | No usable GPUs found
4/14/2016 6:01:44 AM | | Host name: gpugrid

Further down on the log, it shows that GPUGRID is looking for new tasks for CPU (not for GPU!!!) - of course, no such ones can be found.

On the GPUGRID website, where my total 4 hosts are listed with CPU and GPU details, there is no entry for the GPU(s) - the space is left blank.

I closed BOINC and restarted the system twice, but this did not help.

All this is such after I reinstalled BOINC yesterday, with the data directory on a RAMDisk (however, it's been running on a RAMDisk before for several days, so the RAMDisk should not be the cause of this problem).
BTW, the reason for installing BOINC on a RAMDisk was that I am also crunching ATLAS@home (to make better use of my 6-core CPU(HT) and my 32GB RAM), plus this project writes up to 200 GB per day on the SSD (or RAMDisk now).
As said above, for a few days, it's been working fine. Only after I un- and reinstalled BOINC (because there had been an inconsistency which caused a problem with ATLAS, and which could only be solved by reinstalling BOINC), I have this problem with GPUGRID not detecting the two GPUs.

Anyone any idea how I can get the problem solved?

You are of course running VirtualBox in order to run ATLAS. I have had a number of problems with VBox interfering with other programs. In particular, I can't run Folding on a GPU and also run any VBox program (i.e., ATLAS, vLHC, Cosomology) without the Folding client crashing. Folding usually then restarts without an error, but obviously there is some interference.

I have not found problems with VBox interfering with any BOINC GPU projects yet, but it could just be a matter of time. Maybe a new version of VBox will help? (Also, I am on Win7, which provides more isolation of the video drivers than some other versions.)

You are of course running VirtualBox in order to run ATLAS. I have had a number of problems with VBox interfering with other programs.

The only difference in the new setting is that the complete BOINC data directory now is in on the RAMDisc, whereas before only the SLOTS folder was. This, however, turned out not to be good, as ATLAS did not work well.
Now, with the ATLAS problem fixed, the GPUGRID problem exists :-(

I am not sure how you put only the slots folder on the ramdisk

I am not sure how you put only the slots folder on the ramdisk

Actually, I would totally uninstalled VBox if merely stopping ATLAS did not work. I usually search the drive and registry for any "VBox" entries too. If that fixed it, you might then be able to re-install the latest version of VBox and get it to work. It could be that it picked up on the old directory structure, and does not want to let go. I see such things once in a while.

For that matter, you may need to do the same thing with BOINC. When going from a service install to a non-service install, for example, I often have to hack all BOINC references from the registry; also, when changing the DATA directory from a ramdisk back to the SSD, or vice-versa. Once installed, it can be hard to change.

Did you install Boinc as a service?
Did you restart?

No, not running BOINC on a Virtual Operating System.

What I did just now was uninstalling BOINC, cleaning the registry, then re-installing BOINC, then starting the system new, then opening the BOINC manager.
The event log shows good news:

4.04.2016 22:03:41 | | CUDA: NVIDIA GPU 0: GeForce GTX 980 Ti (driver version 361.75, CUDA version 8.0, compute capability 5.2, 4096MB, 4009MB available, 6986 GFLOPS peak)
14.04.2016 22:03:41 | | CUDA: NVIDIA GPU 1: GeForce GTX 980 Ti (driver version 361.75, CUDA version 8.0, compute capability 5.2, 4096MB, 4009MB available, 6986 GFLOPS peak)

However, crunching could not start yet, because:

14.04.2016 22:12:13 | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
14.04.2016 22:12:15 | GPUGRID | Scheduler request completed: got 0 new tasks
14.04.2016 22:12:15 | GPUGRID | No tasks sent
14.04.2016 22:12:15 | GPUGRID | No tasks are available for Long runs (8-12 hours on fastest card)
14.04.2016 22:12:15 | GPUGRID | Project has no tasks available

I definitely would like to see GPUGRID running okay before I add the ATLAS project.
So, hopefully, new GPUGRID WUs will follow soon.

Later last evening, 2 GPUGRID tasks were downloaded and started being crunched, so I attached ATLAS, and ATLAS also worked well.
After the 2 GPUGRID tasks were finished and uploaded, no new ones were downloaded, the log file said:


15.04.2016 05:22:02 | GPUGRID | Sending scheduler request: To fetch work.
15.04.2016 05:22:02 | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
15.04.2016 05:22:04 | GPUGRID | Scheduler request completed: got 0 new tasks
15.04.2016 05:22:04 | GPUGRID | No tasks sent
15.04.2016 05:22:04 | GPUGRID | message from Server: Long runs (8-12 hours on fastest card) needs 2156.64MB more disk space. You currently have 5472.75 MB available and it needs 7629.39 MB.

Indeed, a GPIGRID WU needs that much disk space? If so, my plan to have BOINC's data directory on a 14GB RAMDisk will not work :-( Because I would so have to eliminate some ATLAS tasks, which were actually the reason why I changed to RAMDisk.
Theoretically, I could increase the RAMDisk by 2.5 or so, but by the same amount my system RAM would be cut. Since each ATLAS task uses some 2.5GB RAM, this would mean I could not crunch 4 ATLAS tasks simultaneausly, as its happening now, but maximum 3. To achieve this, I would have to decrease CPU usage setting to 40% - so with my 6-core (+ 6 core HT) CPU I would crunch only 5 tasks simultaneaously. Which is definitely not what I had planned.

Does anyone have any suggestions I might not have thought of?

Edit: It just occurs to me - for having 2 GPUGRID running, I would need even 15.3GB free disc space - more than my total RAMdisk presently is. And I would not be able to ran even one ATLAS task in addition.

I just read a posting in the ATLAS forum where someone else was having a similar issue - and he said that it got solved by installing a separate BOINC client.

I am wonderung whether it is really possible to just setup BOINC twice - this works? No interferences or other complications?
Is there anything I would have to particularly observe during installation or thereafter?

If this could be done, it most probably would solve my problem (unless any other unforeseen issues come up).
I would have one BOINC for ATLAS (with the data directory on the RAMDisk), and another one for GPUGRID (with the data directory on the SSD), and just set the CPU usage %values as needed.

Any comments?

You can try the dual BOINC installation, but there are a number of hoops that you have to jump through. It has never looked worthwhile to me.

I have dodged the bullet on that problem, since it just happens that on the machines where I run ATLAS and GPUGrid, I use PrimoCache rather than a ramdisk. Then, BOINC sees the entire free space of the SSD.

But I wonder if GPUGrid really needs all that much space? I doubt it for the GPU client; it may be an artifact of the CPU work they were doing at one time. Maybe you could post separately about it, and it could be reduced.

However, be aware that GPUGrid doesn't have much work for the moment, since most of their researchers have graduated. A new group is coming in, but I really don't expect much for several months; probably at least the end of the summer.

You can try the dual BOINC installation, but there are a number of hoops that you have to jump through. It has never looked worthwhile to me.

I use PrimoCache rather than a ramdisk.

But I wonder if GPUGrid really needs all that much space? I doubt it for the GPU client; it may be an artifact of the CPU work they were doing at one time. Maybe you could post separately about it, and it could be reduced

However, be aware that GPUGrid doesn't have much work for the moment, since most of their researchers have graduated. A new group is coming in, but I really don't expect much for several months; probably at least the end of the summer.

You can try the dual BOINC installation, but there are a number of hoops that you have to jump through. It has never looked worthwhile to me.

Which ones exactly are the traps?

... and people report different problems ...

Perhaps indeed someone could come up with some script, config.xml or whatever in order to circumvent this strange 7.629MB free disk space requirement.