Hi,

I've installed the nvidia proprietary installed, installed boinc, add projects. steam games are working, proof that the videos drivers are working. So is there a symbolic link that I should do somewhere because boinc doesn't find the right libraries?
OS: fedora 31 workstation

Thanks in advance for all the leads you could provide.

Post the output of
lspci | grep VGA

Did you install the OpenCL drivers for your GPU?

sorry for the delay.

$ lspci | grep VGA
01:00.0 VGA compatible controller: NVIDIA Corporation GM204 [GeForce GTX 970] (rev a1)

And I've isntalled the nvidia proprietary drivers and cuda from nvidia so I guess it should be packaged into that no?
How would I verify that accurately?

$ inxi -GS
System:    Host: fedo-mansion-ada Kernel: 5.3.16-300.fc31.x86_64 x86_64 bits: 64 Desktop: Gnome 3.34.2 
           Distro: Fedora release 31 (Thirty One) 
Graphics:  Device-1: NVIDIA GM204 [GeForce GTX 970] driver: nvidia v: 440.36 
           Display: x11 server: Fedora Project X.org 1.20.6 driver: nvidia resolution: 1920x1080~60Hz, 1920x1080~60Hz 
           OpenGL: renderer: GeForce GTX 970/PCIe/SSE2 v: 4.6.0 NVIDIA 440.36

Did you reboot after installing the drivers?

Usually BOINC will tell you what if found in its log, the first 20 lines or so should show what it found. Please post those log lines.

From memory all it does is look for the OpenCL.so and CUDA libraries in particular folders. Given the 440 drivers are new it’s possible Nvidia put then in a new location.

---

MarkJ

sorry but where would be that log?
Because I don't have anything really consistent with that in /var/lib/boinc
or do you want me to copy paste the event log of boinc manager?

boinc]$ ls
account_setiathome.berkeley.edu.xml     gui_rpc_auth.cfg                        sched_reply_setiathome.berkeley.edu.xml
account_www.worldcommunitygrid.org.xml  job_log_setiathome.berkeley.edu.txt     sched_reply_www.worldcommunitygrid.org.xml
all_projects_list.xml                   job_log_www.worldcommunitygrid.org.txt  sched_request_setiathome.berkeley.edu.xml
client_state_prev.xml                   lockfile                                sched_request_www.worldcommunitygrid.org.xml
client_state.xml                        lookup_account.xml                      slots
coproc_info.xml                         lookup_website.html                     statistics_setiathome.berkeley.edu.xml
daily_xfer_history.xml                  master_setiathome.berkeley.edu.xml      statistics_www.worldcommunitygrid.org.xml
get_current_version.xml                 master_www.worldcommunitygrid.org.xml   stderrgpudetect.txt
get_project_config.xml                  notices                                 stdoutgpudetect.txt
global_prefs.xml                        projects                                time_stats_log

there is nothing really important in the std* txt file and there is no where I can see i nthe documentation the location fo the log for boinc and nothing in /var/log/
Or do you want me to activate some logs flags in the manager?
Best regards

since log on linux for boinc is ont stdoutput so I'm givingyou the first lines of boinc manager event log

Wed 18 Dec 2019 08:59:21 AM CET |  | cc_config.xml not found - using defaults
Wed 18 Dec 2019 08:59:21 AM CET |  | Starting BOINC client version 7.16.1 for x86_64-pc-linux-gnu
Wed 18 Dec 2019 08:59:21 AM CET |  | log flags: file_xfer, sched_ops, task
Wed 18 Dec 2019 08:59:21 AM CET |  | Libraries: libcurl/7.66.0 OpenSSL/1.1.1d-fips zlib/1.2.11 brotli/1.0.7 libidn2/2.3.0 libpsl/0.21.0 (+libidn2/2.2.0) libssh/0.9.3/openssl/zlib nghttp2/1.39.2
Wed 18 Dec 2019 08:59:21 AM CET |  | Data directory: /var/lib/boinc
Wed 18 Dec 2019 08:59:21 AM CET |  | OpenCL CPU: pthread-Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz (OpenCL driver vendor: The pocl project, driver version 1.5-pre, device version OpenCL 1.2 pocl HSTR: pthread-x86_64-unknown-linux-gnu-skylake)
Wed 18 Dec 2019 08:59:21 AM CET |  | No usable GPUs found
Wed 18 Dec 2019 08:59:21 AM CET |  | [libc detection] gathered: 2.30, GNU libc
Wed 18 Dec 2019 08:59:21 AM CET |  | Host name: fedo-mansion-ada
Wed 18 Dec 2019 08:59:21 AM CET |  | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz [Family 6 Model 158 Stepping 9]
Wed 18 Dec 2019 08:59:21 AM CET |  | 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 pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1d
Wed 18 Dec 2019 08:59:21 AM CET |  | OS: Linux Fedora: Fedora 31 (Workstation Edition) [5.3.16-300.fc31.x86_64|libc 2.30 (GNU libc)]
Wed 18 Dec 2019 08:59:21 AM CET |  | Memory: 15.60 GB physical, 7.86 GB virtual
Wed 18 Dec 2019 08:59:21 AM CET |  | Disk: 68.39 GB total, 41.99 GB free

So according to that there aren’t any usable GPUs which suggests you don’t have the right driver installed. It also says you’ve installed the pocl OpenCL which there are reports of it not being suitable for projects to use. I’d suggest getting rid of it.

I use Debian and have to install the nvidia-kernel-dkms package along with nvidia-opencl-icd. Seeing as you’re using Fedora that won’t work for you. A quick google search came up with this: https://rpmfusion.org/Howto/NVIDIA#About_this_Howto which has a section on CUDA. I’m not sure how up to date it is and there is no mention of the RTX2000 or the GTX1600 series of cards (ie current models). You’ll need CUDA and OpenCL to use them on most BOINC projects. There is mention of them being a third party ppa on the Fedora site.

---

MarkJ

I'm installing my drivers manually from the official website. No repo and I'm not really interested to let the control to those repo.
The drivers are well installed as proven before with the different output I've put and because graphical program based on the nvidia drivers wouldn't work otherwise.

But about opencl I don't know since I'm not sure what would be the whole way to install it under fedora. As I've seen there: https://devtalk.nvidia.com/default/topic/1043943/linux/opencl-doesn-t-work-with-nvidia-driver-410-xx-415-13-drivers-for-linux/ The opencl extension are implemented into the nvidia proprietary drivers and the cuda drivers which as stated above I already installed. I can try to reinstall the cuda drivers from the website, not the repo as I did last time for example.

I don't know which version of opencl are available nor how to install it precisely since as shown in the link they are supposed to be in the drivers from nvidia. And the dkms package are already installed and at each update fo the kernel it rebuilds the nvidia driver proprietary as it should do. So the build and use of nvidia drivers here are not in question. They are working perfectly as intended. The problem is where does the boinc-client looking for the librairies?

[root@fedo-mansion-ada ~]# sudo dnf history list all | grep opencl
[root@fedo-mansion-ada ~]# 

I can't get rid of something that is not installed by the system...

[root@fedo-mansion-ada ~]# lsmod | grep opencl
[root@fedo-mansion-ada ~]# 

So I don't know how it detected that something related to opencl is installed.

The only thing I cand find is that:

[root@fedo-mansion-ada ~]# sudo dnf list installed | grep opencl
opencl-filesystem.noarch                           1.0-10.fc31                         @fedora  

well Apparently you were right. There seem to be no support anymore to manually install and activate the cuda requirements without installing the drivers and everything must be through the rpmFusion repo. It bothers me a lot but I will try and gives you feedback when it's done.

Because you have a GTX970 you could use an older driver from Nvidia, not the 440. I found this site that has instructions about installing the official Nvidia drivers which might help.

---

MarkJ

thanks for the link I will look into it.

I've tried the rpmfusion way. Now I have artifacts at boot, steam games that I sued to test the graphic acceleration won't work anymore. I have a less fluid interface now, still harware accelerated but I can detect some differences.
And I'm pretty sure the driver were well installed before because in the uninstaller I've seen some line about libcuda.
And now boinc, which I modify the flag log, is complaining about libcuda. So I'm pretty sure that the system was okey but boinc wasn't detected libcuda before as for now.

Thu 19 Dec 2019 10:01:51 AM CET |  | Starting BOINC client version 7.16.1 for x86_64-pc-linux-gnu
Thu 19 Dec 2019 10:01:51 AM CET |  | log flags: file_xfer, sched_ops, task, coproc_debug, notice_debug
Thu 19 Dec 2019 10:01:51 AM CET |  | Libraries: libcurl/7.66.0 OpenSSL/1.1.1d-fips zlib/1.2.11 brotli/1.0.7 libidn2/2.3.0 libpsl/0.21.0 (+libidn2/2.2.0) libssh/0.9.3/openssl/zlib nghttp2/1.39.2
Thu 19 Dec 2019 10:01:51 AM CET |  | Data directory: /var/lib/boinc
Thu 19 Dec 2019 10:01:51 AM CET |  | [coproc] launching child process at /usr/bin/boinc
Thu 19 Dec 2019 10:01:51 AM CET |  | [coproc] with data directory /var/lib/boinc
Thu 19 Dec 2019 10:01:51 AM CET |  | OpenCL CPU: pthread-Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz (OpenCL driver vendor: The pocl project, driver version 1.5-pre, device version OpenCL 1.2 pocl HSTR: pthread-x86_64-unknown-linux-gnu-skylake)
Thu 19 Dec 2019 10:01:51 AM CET |  | [coproc] NVIDIA: libcuda.so: cannot open shared object file: No such file or directory
Thu 19 Dec 2019 10:01:51 AM CET |  | [coproc] ATI: libaticalrt.so: cannot open shared object file: No such file or directory
Thu 19 Dec 2019 10:01:51 AM CET |  | No usable GPUs found
Thu 19 Dec 2019 10:01:51 AM CET |  | [libc detection] gathered: 2.30, GNU libc
Thu 19 Dec 2019 10:01:51 AM CET |  | Host name: fedo-mansion-ada
Thu 19 Dec 2019 10:01:51 AM CET |  | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz [Family 6 Model 158 Stepping 9]
Thu 19 Dec 2019 10:01:51 AM CET |  | 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 pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1d
Thu 19 Dec 2019 10:01:51 AM CET |  | OS: Linux Fedora: Fedora 31 (Workstation Edition) [5.3.16-300.fc31.x86_64|libc 2.30 (GNU libc)]
Thu 19 Dec 2019 10:01:51 AM CET |  | Memory: 15.60 GB physical, 7.86 GB virtual
Thu 19 Dec 2019 10:01:51 AM CET |  | Disk: 68.39 GB total, 40.75 GB free
Thu 19 Dec 2019 10:01:51 AM CET |  | Local time is UTC +1 hours

So again, where boinc is looking for? Because clearly it's the problem. And it's the presence of libcuda that tells boinc if there is a GPU usable or not, nothing else.

/usr/lib/libcuda.so
/usr/lib/libcuda.so.1
/usr/lib/libcuda.so.440.44
/usr/lib64/libcuda.so
/usr/local/cuda-10.2/doc/man/man7/libcuda.so.7
/usr/local/cuda-10.2/targets/x86_64-linux/lib/stubs/libcuda.so
/usr/share/man/man7/libcuda.so.7

that's where it is installed on my fedora workstation.

Now that I thought about it, using an older driver is not possible when dealing with graphics.
As stated in several steam github issues, the xorg version has to match the driver version so that the usernamespace can access graphic acceleration capabilities.

So again, where boinc is looking for? Because clearly it's the problem. And it's the presence of libcuda that tells boinc if there is a GPU usable or not, nothing else.

https://github.com/BOINC/boinc/blob/master/client/gpu_nvidia.cpp teaches us that BOINC checks for libnvidia-ml.so.1 and libcuda.so using the dlopen function which will check /lib and /usr/lib automatically.

---

okey. Thanks for that information. But since it claims that it doesn't exist, then how?
Should I open an issue then? because the symbolic link is correct so it's not a non-existent file :

$ ls -l /usr/lib/libc*
-rwxr-xr-x. 1 root root  2894668 Dec  4 18:04 /usr/lib/libc-2.30.so
lrwxrwxrwx. 1 root root       29 Jul 24 21:51 /usr/lib/libcairo-gobject.so.2 -> libcairo-gobject.so.2.11600.0
-rwxr-xr-x. 1 root root    35680 Jul 24 21:51 /usr/lib/libcairo-gobject.so.2.11600.0
lrwxrwxrwx. 1 root root       40 Jul 24 21:51 /usr/lib/libcairo-script-interpreter.so.2 -> libcairo-script-interpreter.so.2.11600.0
-rwxr-xr-x. 1 root root   155976 Jul 24 21:51 /usr/lib/libcairo-script-interpreter.so.2.11600.0
lrwxrwxrwx. 1 root root       21 Jul 24 21:51 /usr/lib/libcairo.so.2 -> libcairo.so.2.11600.0
-rwxr-xr-x. 1 root root  1404672 Jul 24 21:51 /usr/lib/libcairo.so.2.11600.0
lrwxrwxrwx. 1 root root       14 Jul 25 22:12 /usr/lib/libcap.so.2 -> libcap.so.2.26
-rwxr-xr-x. 1 root root    28828 Jul 25 22:12 /usr/lib/libcap.so.2.26
-rwxr-xr-x. 1 root root    19576 Jul 24 22:47 /usr/lib/libcolordcompat.so
lrwxrwxrwx. 1 root root       25 Jul 24 22:46 /usr/lib/libcolordprivate.so.2 -> libcolordprivate.so.2.0.5
-rwxr-xr-x. 1 root root   253600 Jul 24 22:47 /usr/lib/libcolordprivate.so.2.0.5
lrwxrwxrwx. 1 root root       18 Jul 24 22:46 /usr/lib/libcolord.so.2 -> libcolord.so.2.0.5
-rwxr-xr-x. 1 root root   416016 Jul 24 22:47 /usr/lib/libcolord.so.2.0.5
lrwxrwxrwx. 1 root root       20 Jul 24 22:46 /usr/lib/libcolorhug.so.2 -> libcolorhug.so.2.0.5
-rwxr-xr-x. 1 root root   107932 Jul 24 22:47 /usr/lib/libcolorhug.so.2.0.5
lrwxrwxrwx. 1 root root       17 Aug  5 16:18 /usr/lib/libcom_err.so.2 -> libcom_err.so.2.1
-rwxr-xr-x. 1 root root    21264 Aug  5 16:18 /usr/lib/libcom_err.so.2.1
lrwxrwxrwx. 1 root root       19 Oct  3 18:42 /usr/lib/libcrypto.so.1.1 -> libcrypto.so.1.1.1d
-rwxr-xr-x. 1 root root  3025908 Oct  3 18:42 /usr/lib/libcrypto.so.1.1.1d
lrwxrwxrwx. 1 root root       17 Dec 15 21:56 /usr/lib/libcrypt.so.2 -> libcrypt.so.2.0.0
-rwxr-xr-x. 1 root root   221940 Dec 15 21:56 /usr/lib/libcrypt.so.2.0.0
lrwxrwxrwx. 1 root root       12 Dec  4 17:53 /usr/lib/libc.so.6 -> libc-2.30.so
lrwxrwxrwx. 1 root root       17 Dec 11 21:02 /usr/lib/libcuda.so -> libcuda.so.440.44
lrwxrwxrwx. 1 root root       17 Dec 11 21:02 /usr/lib/libcuda.so.1 -> libcuda.so.440.44
-rwxr-xr-x. 1 root root 15584352 Dec  8 04:38 /usr/lib/libcuda.so.440.44
-rwxr-xr-x. 1 root root    44388 Oct 16 16:05 /usr/lib/libcupsimage.so.2
-rwxr-xr-x. 1 root root   659208 Oct 16 16:05 /usr/lib/libcups.so.2
lrwxrwxrwx. 1 root root       16 Sep 13 10:27 /usr/lib/libcurl.so.4 -> libcurl.so.4.6.0
-rwxr-xr-x. 1 root root   682656 Sep 13 10:27 /usr/lib/libcurl.so.4.6.0

libcuda.so only checks for which CUDA version you have, it doesn't check which Nvidia GPU you have, which is a first requirement done by libnvidia-ml.so.1

---

you are avoiding the question because the error message specifies that libcuda is not found and I just show you that it is there and that the symbolic link is functional.

But okey let's talk about the other lib then:

$ sudo locate libnvidia-ml.so.1
/usr/lib/libnvidia-ml.so.1

$ sudo ls -l /usr/lib/libnvi*
lrwxrwxrwx. 1 root root       29 Dec 19 09:57 /usr/lib/libnvidia-allocator.so.1 -> libnvidia-allocator.so.440.44
-rwxr-xr-x. 1 root root    76136 Dec  8 04:24 /usr/lib/libnvidia-allocator.so.440.44
-rwxr-xr-x. 1 root root 50752496 Dec  8 04:38 /usr/lib/libnvidia-compiler.so.440.44
-rwxr-xr-x. 1 root root 26061720 Dec  8 04:39 /usr/lib/libnvidia-eglcore.so.440.44
lrwxrwxrwx. 1 root root       30 Dec 19 09:57 /usr/lib/libnvidia-egl-wayland.so.1 -> libnvidia-egl-wayland.so.1.1.4
-rwxr-xr-x. 1 root root    55076 Sep 15 11:59 /usr/lib/libnvidia-egl-wayland.so.1.1.4
lrwxrwxrwx. 1 root root       26 Dec 11 21:02 /usr/lib/libnvidia-encode.so.1 -> libnvidia-encode.so.440.44
-rwxr-xr-x. 1 root root    96800 Dec  8 04:24 /usr/lib/libnvidia-encode.so.440.44
-rwxr-xr-x. 1 root root   405064 Dec  8 04:25 /usr/lib/libnvidia-fatbinaryloader.so.440.44
lrwxrwxrwx. 1 root root       23 Dec 19 09:57 /usr/lib/libnvidia-fbc.so.1 -> libnvidia-fbc.so.440.44
-rwxr-xr-x. 1 root root   125892 Dec  8 04:24 /usr/lib/libnvidia-fbc.so.440.44
-rwxr-xr-x. 1 root root 29348904 Dec  8 04:38 /usr/lib/libnvidia-glcore.so.440.44
-rwxr-xr-x. 1 root root   633964 Dec  8 04:25 /usr/lib/libnvidia-glsi.so.440.44
-rwxr-xr-x. 1 root root 11556076 Dec  8 04:37 /usr/lib/libnvidia-glvkspirv.so.440.44
lrwxrwxrwx. 1 root root       23 Dec 19 09:57 /usr/lib/libnvidia-ifr.so.1 -> libnvidia-ifr.so.440.44
-rwxr-xr-x. 1 root root   192000 Dec  8 04:24 /usr/lib/libnvidia-ifr.so.440.44
lrwxrwxrwx. 1 root root       22 Dec 11 21:02 /usr/lib/libnvidia-ml.so.1 -> libnvidia-ml.so.440.44
-rwxr-xr-x. 1 root root  1558556 Dec  8 04:26 /usr/lib/libnvidia-ml.so.440.44
lrwxrwxrwx. 1 root root       26 Dec 11 21:02 /usr/lib/libnvidia-opencl.so.1 -> libnvidia-opencl.so.440.44
-rwxr-xr-x. 1 root root 25884712 Dec  8 04:38 /usr/lib/libnvidia-opencl.so.440.44
lrwxrwxrwx. 1 root root       31 Dec 11 21:02 /usr/lib/libnvidia-opticalflow.so.1 -> libnvidia-opticalflow.so.440.44
-rwxr-xr-x. 1 root root    35264 Dec  8 04:24 /usr/lib/libnvidia-opticalflow.so.440.44
lrwxrwxrwx. 1 root root       34 Dec 11 21:02 /usr/lib/libnvidia-ptxjitcompiler.so.1 -> libnvidia-ptxjitcompiler.so.440.44
-rwxr-xr-x. 1 root root  9086920 Dec  8 04:29 /usr/lib/libnvidia-ptxjitcompiler.so.440.44
-rwxr-xr-x. 1 root root    13400 Dec  8 04:24 /usr/lib/libnvidia-tls.so.440.44

it is there also. And to my knowledge GTX970 is totally compatible CUDA.

so it's totally not a problem of lib since they are there and that the hardware acceleration is working and the cuda application of nvidia are working as well.

So what should I do ? a ticket then?

Do you have either of

stderrgpudetect.txt
stdoutgpudetect.txt

in your boinc-client directory? Do they help?

Please add the following cc_config.xml file to your BOINC data directory /var/lib/boinc

<cc_config>
<log_flags>
<coproc_debug>1</coproc_debug>
</log_flags>
<options>
</options>
</cc_config>

If the file exists, add the line in the correct place, or change its 0 to 1.

After that exit & restart the BOINC client.
Then post the output of the event log again.

---