"No usable GPUs found"

Sorry. I am sure this has been discussed but ....

I have encountered this several times on Linux systems and I have been able to "guess" why the BOINC client is not able to find the GPU. I have browsed through the client source a couple times to glean more information about the testing being done to arrive at a FOUND GPU.

I use the standard Linux distribution version of the manager/client. Is there a way to get more detailed information about why the BOINC client puts the GPU in the not found category?

Do you know of a method displaying that information or a "diagnostic tool" that would help repair the upgrade?

The startup diagnostics are GREAT. I was using Ubuntu 14.04 LTS and the GPU was found and used by projects. An automatic upgrade to 14.10 now gets the NOT FOUND message.


Thanks
rod


Fri 08 May 2015 04:08:55 PM PDT | | Starting BOINC client version 7.4.8 for x86_64-pc-linux-gnu
Fri 08 May 2015 04:08:55 PM PDT | | log flags: file_xfer, sched_ops, task
Fri 08 May 2015 04:08:55 PM PDT | | Libraries: libcurl/7.37.1 OpenSSL/1.0.1f zlib/1.2.8 libidn/1.28 librtmp/2.3
Fri 08 May 2015 04:08:55 PM PDT | | Data directory: /var/lib/boinc-client
Fri 08 May 2015 04:08:55 PM PDT | | No usable GPUs found
Fri 08 May 2015 04:08:55 PM PDT | | Host name: <host>
Fri 08 May 2015 04:08:55 PM PDT | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
Fri 08 May 2015 04:08:55 PM PDT | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 popcnt tsc_deadline_timer aes xsave avx lahf_lm ida arat epb xsaveopt pln pts dtherm tpr_shadow vnmi flexpriority ept vpid
Fri 08 May 2015 04:08:55 PM PDT | | OS: Linux: 3.16.0-37-generic
Fri 08 May 2015 04:08:55 PM PDT | | Memory: 7.78 GB physical, 7.98 GB virtual
Fri 08 May 2015 04:08:55 PM PDT | | Disk: 578.83 GB total, 511.18 GB free
Fri 08 May 2015 04:08:55 PM PDT | | Local time is UTC -7 hours

For BOINC to be able to find the GPU, the GPU needs to be able to do CUDA in the case of an Nvidia GPU, or OpenCL in the case of others. For this you need drivers installed, and possibly added libraries, depending on the distro.

If you have an Intel GPU (like ones built into the Intel CPU) you're out of luck as Intel still hasn't released any OpenCL capable drivers for their GPUs for Linux.

Nvidia and AMD drivers can be downloaded from the Nvidia and AMD websites, or gotten through your distro's repository. Do also look through the sticky thread in this forum. It may be old and not everything may work, but it may give you some clues.

---

Please check if you have the latest versions of boinc-amd-opencl or boinc-nvidia-cuda installed. This should install any missing package that prevents the detection.

Thanks, ChristianB.

I had a working version of Ubuntu 14.04, both CPU and GPU. I "automatically" upgraded to 14.10 and the CPU continued to work fine. The GPU went missing. I can fix it but my effort would be much shorter if I knew WHAT the BOINC client could not find.

I had "incorrectly" put this message on the BOINC CLIENT WISH LIST board to request that the BOINC client display the status of events that prevent the GPU from being found. The moderator thought I was requesting general installation help and moved the thread here.

The BOINC client software looks for two dynamic libraries: libnvidia-ml.so and libcuda.so. Both are currently installed on my system. The 14.10 upgrade just goofed up one of them. If the client cannot find them, that is one problem. If there is a version conflict or access rights problem that is another.

If the BOINC client displayed a "status code" and told me why my configuration is sick, that would help. If there was a client option that said: "test the GPU now" rather than having to start/stop ... that would help too.

The BOINC client knows what the problem is. I was requesting that it share that information. 8-)


handle = dlopen("libnvidia-ml.so", RTLD_NOW);

#ifdef __APPLE__
cudalib = dlopen("/usr/local/cuda/lib/libcuda.dylib", RTLD_NOW);
#else
cudalib = dlopen("libcuda.so", RTLD_NOW);
#endif

BOINC can tell you all sort of things but you need to mutter the right magical incantation, <coproc_debug> in this case.

I'm going to guess that BOINC, or rather the driver, will tell you that BOINC has no permission to access the GPU. Installing nvidia-modprobe should help with that. You may need to enable backports repository to get the package.

If you want to run OpenCL science apps you may need to add a symlink libOpenCL.so -> libOpenCL.so.1 . That was removed from driver packages some versions ago. Or alternatively install ocl-icd-libopencl1 which still includes the symlink.


And btw, Intel GPUs can be used under Linux these days with Beignet, although not every combination of GPU+driver+kernel+science app works. And of course the GPU need to support OpenCL to begin with. I don't know if the GPU in i7-2600 is OpenCL compatible or not. Beignet development is done by Intel employees so I think it's ok to say Intel has released it.

Thanks much Juha,
The BOINC EXPERTS and VOLUNTEERS have been great.
Your information about the symlink to libOpenCL.so will help me. I thought of setting the ADVANCED -> "Event log diagnostics" -> coproc_debug flag but the Ubuntu manager version predates that feature. I did not think of adding to the cc_config.xml file (DOH!). Upon searching the boinc client 7.6.5.6 tarball, I did not see where it displays the library version information for the user.

I just gave up and REINSTALLED Ubuntu 14.04 OVER the top of my 14.10 upgrade and everything came back working PERFECTLY again. It took me less frustration to surrender and revert to 14.04 than to try to debug the new 14.10 BOINC manager/client system.

I was originally suggesting a BOINC CLIENT STARTUP ... AUTO-MUTTER feature. I have read the "magical incantation" page many times. HOWEVER, a list of required libraries and paths that BOINC manager/client searches to "FOUND" a "USABLE GPU" on the "magical incantation" page would be EVEN BETTER than my suggestion.

It is pretty easy to munch up a Linux distribution when installing this stuff and get multiple copies of libraries (like I did) and ... not easy to figure out the problem.

thanks again