I have always had this message, but I recently upgraded my video card to a Radeon HD6750 and according to documentation on this site, this GPU is usable. So how do I make it happen ?

I'm running Linux Fedora 19 and installed Boinc from the repositories so boinc is happily running from /var/lib/boinc

I really don't want to use the Berkley BOINC installer because that created all sorts of problems for me in the past.

I have tried all of this ...http://boinc.berkeley.edu/dev/forum_thread.php?id=8129#47188 to no avail.

I tried putting symlinks in /var/lib/boinc to the various libraries - though I couldn't see why that would work - and it didn't.

Any further suggestions ?

Have you tried the fixes in the following thread?:

Debian/Ubuntu/Mint/Derivatives - GPU recognition fixes

Claggy

OK. Tried all that, it doesn't work.

Fedora is not exactly the same as Debian/Ubuntu in the way it starts things - it's running systemd, and doesn't use /etc/init.d but I went through the same processes of starting/restarting the boinc-client after making changes, and it affected nothing. Still got ... No usable GPUs found

OK. Got a bit further with this, the GPU is now recognised. What exactly makes the difference I don't know, so some of the following may not be necessary.

The set up is Fedora 19 with the AMD Catalyst Beta 13.8 driver.

I have NOT disabled Selinux. Some users advocate it - it isn't necessary.
I have NOT modified the firewall.

I added boinc to the video group.

There's a bug in the 13.8 beta driver which means access rights to the device are not set correctly and so for example the AMD Catalyst Control Centre program does not start unless you are root. We don't like being root, so you need to enable access to /proc/ati with a chmod -R 777 /proc/ati

To allow this to happen, and to make sure the video drivers are there before boinc starts I amended /usr/lib/systemd/system/boinc-client.service with the following lines...

ExecStartPre=/bin/sleep 20
ExecStartPre=/usr/bin/chmod -R 777 /proc/ati

i.e. changed the sleep from 1 to 20, and added the line to change permissions.

I also added the following line to allow the Xserver realise that boinc is allowed access to the GPUs

ExecStartPre=/usr/bin/xhost local:boinc

Boinc now starts up and says I have a GPU ...

Mon 26 Aug 2013 16:49:08 BST | | CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 964MB available, 50 GFLOPS peak)

So, there it is. Whether it's actually being used is another question.

Hmmm, I just noticed it says I have a HD 5700 series. But ...

[mike@ceri ~]$ lspci | grep VGA
02:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Juniper PRO [Radeon HD 6750]
[mike@ceri ~]$

One of lifes' little mysteries I guess

Mon 26 Aug 2013 16:49:08 BST | | CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 964MB available, 50 GFLOPS peak)

Hmmm, I just noticed it says I have a HD 5700 series. But ...

[mike@ceri ~]$ lspci | grep VGA
02:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Juniper PRO [Radeon HD 6750]

That's because the detection code in BOINC only sees the HD5700 as a Juniper. I sent in code to the developers that adds the HD6750 and HD6770 as well. That'll be added in a future version of BOINC.

---

Hmm, well something is still not right, because I'm not getting work units. Here's a log. Einstein thinks the driver doesn't support OpenCL ... or I've set things up incorrectly.

2013-08-27 11:03:56.9038 [PID=7385]   Request: [USER#xxxxx] [HOST#xxxxxxxx] [IP xxx.xxx.xxx.16] client 7.0.65
2013-08-27 11:03:56.9159 [PID=7385 ]    [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2013-08-27 11:03:56.9159 [PID=7385 ]    [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] Not using matchmaker scheduling; Not using EDF sim
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] ATI: req 108000.00 sec, 1.00 instances; est delay 0.00
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] work_req_seconds: 0.00 secs
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] available disk 5.79 GB, work_buf_min 64800
2013-08-27 11:03:56.9160 [PID=7385 ]    [send] active_frac 0.999767 on_frac 0.285188 DCF 1.199328
2013-08-27 11:03:56.9171 [PID=7385 ]    [send] [HOST#8311447] is reliable
2013-08-27 11:03:56.9171 [PID=7385 ]    [send] set_trust: random choice for error rate 0.027737: no
2013-08-27 11:03:56.9172 [PID=7385 ]    [mixed] sending locality work first (0.1220)
2013-08-27 11:03:57.2205 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 105 for einstein_S6CasA ()
2013-08-27 11:03:57.2206 [PID=7385 ]    [version] Checking plan class 'X64'
2013-08-27 11:03:57.2214 [PID=7385 ]    [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
2013-08-27 11:03:57.2214 [PID=7385 ]    [version] plan class ok
2013-08-27 11:03:57.2214 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 105 for einstein_S6CasA (X64)
2013-08-27 11:03:57.2215 [PID=7385 ]    [version] Checking plan class 'SSE2'
2013-08-27 11:03:57.2215 [PID=7385 ]    [version] plan class ok
2013-08-27 11:03:57.2215 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 105 for einstein_S6CasA (SSE2)
2013-08-27 11:03:57.2215 [PID=7385 ]    [version] no app version available: APP#24 (einstein_S6CasA) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2215 [PID=7385 ]    [version] no app version available: APP#24 (einstein_S6CasA) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2220 [PID=7385 ]    [mixed] sending non-locality work second
2013-08-27 11:03:57.2367 [PID=7385 ]    [version] Checking plan class 'BRP5-cuda32-nv270'
2013-08-27 11:03:57.2368 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2368 [PID=7385 ]    [version] No CUDA devices found
2013-08-27 11:03:57.2368 [PID=7385 ]    [version] Checking plan class 'BRP5-opencl-ati'
2013-08-27 11:03:57.2368 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] ATI device (or driver) doesn't support OpenCL
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] Checking plan class 'BRP5-cuda32-nv270'
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] No CUDA devices found
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] Checking plan class 'BRP5-opencl-ati'
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] ATI device (or driver) doesn't support OpenCL
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] no app version available: APP#23 (einsteinbinary_BRP5) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] no app version available: APP#23 (einsteinbinary_BRP5) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2369 [PID=7385 ]    [version] Checking plan class 'BRP4G-cuda32-nv270'
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] No CUDA devices found
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] Checking plan class 'BRP4G-opencl-ati'
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] ATI device (or driver) doesn't support OpenCL
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] Checking plan class 'BRP4G-cuda32-nv270'
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] No CUDA devices found
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] Checking plan class 'BRP4G-opencl-ati'
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] parsed project prefs setting 'gpu_util_brp': 0.000000
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] ATI device (or driver) doesn't support OpenCL
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] no app version available: APP#25 (einsteinbinary_BRP4G) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2370 [PID=7385 ]    [version] no app version available: APP#25 (einsteinbinary_BRP4G) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] Checking plan class 'BRP4X64'
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] plan class ok
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 133 for einsteinbinary_BRP4 (BRP4X64)
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] Checking plan class 'BRP4SSE'
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] plan class ok
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 134 for einsteinbinary_BRP4 (BRP4SSE)
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2371 [PID=7385 ]    [version] Don't need CPU jobs, skipping version 110 for hsgamma_FGRP2 ()
2013-08-27 11:03:57.2372 [PID=7385 ]    [version] no app version available: APP#21 (hsgamma_FGRP2) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2372 [PID=7385 ]    [version] no app version available: APP#21 (hsgamma_FGRP2) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2013-08-27 11:03:57.2430 [PID=7385 ] [debug]   [HOST#8311447] MSG(high) No work sent
2013-08-27 11:03:57.2430 [PID=7385 ] [debug]   [HOST#8311447] MSG(high) see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/8311/8311447
2013-08-27 11:03:57.2430 [PID=7385 ]    Sending reply to [HOST#8311447]: 0 results, delay req 60.00
2013-08-27 11:03:57.2433 [PID=7385 ]    Scheduler ran 0.346 seconds

[/code][/list]

Hmm, well something is still not right, because I'm not getting work units. Here's a log. Einstein thinks the driver doesn't support OpenCL ... or I've set things up incorrectly.

Well, that's correct when the only thing in your start up log is the CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 964MB available, 50 GFLOPS peak) line. The CAL in front of it shows that the only detection BOINC could do was of the older CAL format.
The OpenCL detection will have a separate line, like so (from my machine):

8/27/2013 3:42:05 AM | | CAL: ATI GPU 0: AMD Radeon HD 7850/7870 series (Pitcairn) (CAL version 1.4.1741, 2048MB, 2008MB available, 6400 GFLOPS peak)
8/27/2013 3:42:05 AM | | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 7850/7870 series (Pitcairn) (driver version 1214.3 (VM), device version OpenCL 1.2 AMD-APP (1214.3), 2048MB, 2008MB available, 6400 GFLOPS peak)

So the AMD drivers should contain the OpenCL part and BOINC should be able to read the correct library, in the case of Linux libOpenCL.so

---

OK. Thanks for clearing that up.

According to the AMD site, the driver can do OpenCL. I've got some Open CL libraries installed with the same date as when I installed the driver, so I guess all I have to do now is work out why boinc can't see/use them

OK, I give in. I need help with this.

Boinc detects my GPU OK. But it doesn't realise it can do OpenCL. The relevant lines of the startup log are ...

30-Aug-2013 15:26:57 [---] Starting BOINC client version 7.0.65 for x86_64-pc-linux-gnu
30-Aug-2013 15:26:57 [---] log flags: file_xfer, sched_ops, task
30-Aug-2013 15:26:57 [---] Libraries: libcurl/7.29.0 NSS/3.14.3.0 zlib/1.2.7 libidn/1.26 libssh2/1.4.3
30-Aug-2013 15:26:57 [---] Data directory: /var/lib/boinc
30-Aug-2013 15:26:57 [---] Processor: 4 AuthenticAMD AMD Athlon(tm) II X4 635 Processor [Family 16 Model 5 Stepping 2]
30-Aug-2013 15:26:57 [---] Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm 3dnowext 3dnow constant_tsc rep_good nopl nonstop_tsc extd_apicid pni monitor cx16 popcnt lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt hw_pstate npt lbrv svm_lock nrip_save
30-Aug-2013 15:26:57 [---] OS: Linux: 3.10.7-200.fc19.x86_64
30-Aug-2013 15:26:57 [---] Memory: 3.86 GB physical, 3.89 GB virtual
30-Aug-2013 15:26:57 [---] Disk: 49.09 GB total, 13.80 GB free
30-Aug-2013 15:26:57 [---] Local time is UTC +1 hours
30-Aug-2013 15:26:57 [---] CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 936MB available, 50 GFLOPS peak)
30-Aug-2013 15:26:57 [climateprediction.net] URL http://climateprediction.net/; Computer ID 1186128; resource share 100
30-Aug-2013 15:26:57 [SETI@home] URL http://setiathome.berkeley.edu/; Computer ID 7061073; resource share 100
30-Aug-2013 15:26:57 [Einstein@Home] URL http://einstein.phys.uwm.edu/; Computer ID 8311447; resource share 100

I have OpenCL 1.2 libraries. "clinfo" tells me that.

[mike@ceri monitor]$ clinfo
Number of platforms: 1
Platform Profile: FULL_PROFILE
Platform Version: OpenCL 1.2 AMD-APP (1272.2)
Platform Name: AMD Accelerated Parallel Processing
Platform Vendor: Advanced Micro Devices, Inc.
Platform Extensions: cl_khr_icd cl_amd_event_callback cl_amd_offline_devices

I understand that boinc cannot look outside its' environment, so in /var/lib/boinc, I've set up some symbolic links.

[mike@ceri monitor]$ ls -al /var/lib/boinc/lib*
lrwxrwxrwx. 1 root root 19 Aug 30 15:25 /var/lib/boinc/libaticalrt.so -> /lib/libaticalrt.so
lrwxrwxrwx. 1 root root 19 Aug 30 15:26 /var/lib/boinc/libOpenCL.so -> /lib/libOpenCL.so.1
[mike@ceri monitor]$ 

I've also tried pointing to the 64-bit libraries of the same name.

boinc doesn't detect that my drivers can do OpenCL.

What else do I need to do ?

Thanks

I did a search through the forums and found this post which you may try to follow. That's for Fedora 17, but I just assume Fedora 19 to work the same way.

Apropos, you did install the videocard drivers from AMD? Although, of course, according to their release notes your OS isn't supported. ;-)

Now, do know you're not the only one with problems. In this thread, jay-e has a lot of head scratches. I'm not sure if he finally manages to get his GPU to be detected.

---

I did a search through the forums and found this post which you may try to follow. That's for Fedora 17, but I just assume Fedora 19 to work the same way.

Unfortunately not. Fedora 17 used the "normal" /etc/init.d stuff. Fedora 19 has gone to using systemd. That thread did lead me to the "xhost local:boinc" statement though, and for some reason it does not work with systemd. The command is there, but it fails "no protocol specified" - the error never comes out on a logfile anywhere - I only discovered it by putting in a lot of status and error checking code. To be fair it doesn't make much real sense to have a boot time service depending on using graphics drivers, when there may not be a graphics card at all in the system. That said, if after the system is up and running I manually do a "xhost local:boinc" command, and then restart the boinc-client service, boinc does see my GPU. It just doesn't see any of the OpenCL stuff.

My current thinking is to set up the LD_LIBRARY_PATH before the boinc starts and see if that makes a difference.

Apropos, you did install the videocard drivers from AMD? Although, of course, according to their release notes your OS isn't supported. ;-)

Now, do know you're not the only one with problems. In this thread, jay-e has a lot of head scratches. I'm not sure if he finally manages to get his GPU to be detected.

Err, yeah. I think a part of the problem is that Fedora is always cutting edge. It is now running kernel 3.10.9-200, and the stable ATI driver does not compile/install. I have to use the beta driver. There are known bugs in it with privileges on the /proc/ati device. Maybe there are other things I've not detected yet with this beta driver.

So, my current state is, the GPU is detected, but not the OpenCL

Mike

OK. Got a bit further with this (again). In fact I've SOLVED it.

By enabling the <coproc_debug>1</coproc_debug> flag in cc_config.xml I got a few more messages in boinc.log.

31-Aug-2013 15:36:34 [---] Local time is UTC +1 hours
31-Aug-2013 15:36:34 [---] CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 955MB available, 50 GFLOPS peak)
31-Aug-2013 15:36:34 [---] No NVIDIA library found
31-Aug-2013 15:36:34 [---] No OpenCL library found
31-Aug-2013 15:36:34 [climateprediction.net] URL http://climateprediction.net/; Computer ID 1186128; resource share 100

So, by looking at the source code for boinc, the "No OpenCL library found" text only appears twice, and both occurrences are in ./client/gpu_opencl.cpp, once for Windows and once for everything else that isn't Apple - so that's me, and the code looks like ...

#else
//TODO: Is this correct?
    opencl_lib = dlopen("libOpenCL.so", RTLD_NOW);
#endif
    if (!opencl_lib) {
        warnings.push_back("No OpenCL library found");
        return;
    }

I built the source code I had for boinc from git, and added a call to dlerror() to get the particular error. I got this ...

Aug 31 18:46:12 ceri boinc[25731]: 31-Aug-2013 18:46:12 [---] libOpenCL.so: cannot open shared object file: No such file or directory. No OpenCL library found

To after a bit more debugging. I discovered that boinc DOES NOT LOOK in /var/lib/boinc for the ATI and OpenCL libraries. It really uses LD_LIBRARY_PATH.

Amend /usr/lib/systemd/system/boinc-client.service to include the line.

Environment="LD_LIBRARY_PATH=/usr/lib64:/lib64:/usr/lib:/lib:/usr/local/lib"
... or where ever your libraries are.

Then make sure you library is libOpenCL.so and not - as was my case libOpenCL.so.1. There's something odd with symlinks here - it appears as if symlinks are ignored by dlopen(). You need a real library without the version number

And it will burst into life.

31-Aug-2013 19:06:13 [---] CAL: ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (CAL version 1.4.1848, 1024MB, 952MB available, 50 GFLOPS peak)
31-Aug-2013 19:06:13 [---] OpenCL: AMD/ATI GPU 0: ATI Radeon HD 5700 series (Juniper) (driver version 1272.2, device version OpenCL 1.2 AMD-APP (1272.2), 1024MB, 952MB available, 50 GFLOPS peak)
3


Yippee !

Great! If you make a post about that in the GPU forum, I'll sticky it.

---