Hey all,

I recently got an Radeon RX 560 graphics card, hoping that it would boost the performance of BOINC, but it doesn't seem as though the card is being detected by the client. The Event Log says: "No usable GPUs found." I'm running Ubuntu 18.04 with BOINC 7.9.3 and the open source AMD graphics drivers. Does anyone have any ideas as to why this is? Is it because I'm not running the proprietary AMD drivers, or is there a setting somewhere that I'm missing?

Here's the output of lspci | grep VGA:

01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Baffin [Radeon RX 550 640SP / RX 560/560X] (rev cf)

And some of the Event Log:

Sun 06 May 2018 09:28:30 AM EDT |  | Starting BOINC client version 7.9.3 for x86_64-pc-linux-gnu
Sun 06 May 2018 09:28:30 AM EDT |  | log flags: file_xfer, sched_ops, task
Sun 06 May 2018 09:28:30 AM EDT |  | Libraries: libcurl/7.58.0 OpenSSL/1.1.0g zlib/1.2.11 libidn2/2.0.4 libpsl/0.19.1 (+libidn2/2.0.4) nghttp2/1.30.0 librtmp/2.3
Sun 06 May 2018 09:28:30 AM EDT |  | Data directory: /var/lib/boinc-client
Sun 06 May 2018 09:28:30 AM EDT |  | No usable GPUs found
Sun 06 May 2018 09:28:30 AM EDT |  | [libc detection] gathered: 2.27, Ubuntu GLIBC 2.27-3ubuntu1
Sun 06 May 2018 09:28:30 AM EDT |  | Host name: god-emperor
Sun 06 May 2018 09:28:30 AM EDT |  | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz [Family 6 Model 60 Stepping 3]
Sun 06 May 2018 09:28:30 AM EDT |  | 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 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 aes xsave avx f16c rdrand lahf_lm abm cpuid_fault epb invpcid_single pti tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm xsaveopt dtherm ida arat pln pts
Sun 06 May 2018 09:28:30 AM EDT |  | OS: Linux Ubuntu: Ubuntu 18.04 LTS [4.15.0-20-generic|libc 2.27 (Ubuntu GLIBC 2.27-3ubuntu1)]
Sun 06 May 2018 09:28:30 AM EDT |  | Memory: 15.61 GB physical, 2.00 GB virtual
Sun 06 May 2018 09:28:30 AM EDT |  | Disk: 457.35 GB total, 421.16 GB free
Sun 06 May 2018 09:28:30 AM EDT |  | Local time is UTC -4 hours
Sun 06 May 2018 09:28:30 AM EDT |  | Config: GUI RPCs allowed from:
Sun 06 May 2018 09:28:30 AM EDT | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 8506104; resource share 100
Sun 06 May 2018 09:28:30 AM EDT |  | No general preferences found - using defaults
Sun 06 May 2018 09:28:30 AM EDT |  | Reading preferences override file
Sun 06 May 2018 09:28:30 AM EDT |  | Preferences:
Sun 06 May 2018 09:28:30 AM EDT |  | max memory usage when active: 7991.26 MB
Sun 06 May 2018 09:28:30 AM EDT |  | max memory usage when idle: 14384.26 MB
Sun 06 May 2018 09:28:30 AM EDT |  | max disk usage: 411.62 GB
Sun 06 May 2018 09:28:30 AM EDT |  | don't compute while active
Sun 06 May 2018 09:28:30 AM EDT |  | don't use GPU while active
Sun 06 May 2018 09:28:30 AM EDT |  | suspend work if non-BOINC CPU load exceeds 25%
Sun 06 May 2018 09:28:30 AM EDT |  | (to change preferences, visit a project web site or select Preferences in the Manager)
Sun 06 May 2018 09:28:30 AM EDT |  | Setting up project and slot directories
Sun 06 May 2018 09:28:30 AM EDT |  | Checking active tasks
Sun 06 May 2018 09:28:30 AM EDT |  | Setting up GUI RPC socket
Sun 06 May 2018 09:28:30 AM EDT |  | gui_rpc_auth.cfg is empty - no GUI RPC password protection
Sun 06 May 2018 09:28:30 AM EDT |  | Checking presence of 105 project file

TIA

and the open source AMD graphics drivers

Yes, and as far as I can find, they're just that, 2D and 3D graphics card drivers. They don't have the necessary component for GPGPUs: OpenCL. AMD's AMDGPU-PRO driver does have OpenCL built in. This is also the correct OpenCL, don't use Mesa or other stuff, as those don't work with the OpenCL used in the projects either.

---

I believe AMDGPU-PRO hasn't yet been compatible with Ubuntu 18.04, but I don't know the hottest news about that situation.

I noticed that AMD does have a pre-release of its Radeon™ Software for Linux for Ubuntu 18.04: https://support.amd.com/en-us/kb-articles/Pages/Radeon-Software-for-Linux-18.20-Early-Preview-Release-Notes.aspx

Is this different than the AMDGPU-PRO driver?

Looks like that package includes both "AMDGPU All-Open" and "AMDGPU-Pro Driver".
User can choose which one to install. "Pro Driver" should be good for RX 560.

I'm not sure if the 'pal' or 'legacy' OpenCL option is the right one for RX 560, but instructions say it's possible to install even both:

./amdgpu-pro-install -y --opencl=legacy,pal

I believe AMDGPU-PRO hasn't yet been compatible with Ubuntu 18.04, but I don't know the hottest news about that situation.

The best place for recent news is probably phoronix.com.

AMDGPU-PRO has been deprecated as a name and now you need to look for "Radeon Software for Linux". This includes the open source stuff, the -Pro stuff and the ROCm (Radeon Open Compute) stuff. There was a new release of Radeon Software for Linux in late April (version 18.10) and the release notes state that Ubuntu 18.04 is not yet supported. However, apparently that is all about to change. There is already on phoronix, a discussion about the pre-release of version 18.20 which now supports Ubu 18.04. I don't use Ubuntu so I don't know any details. I suggest the OP should go read that discussion.

I run a lot of AMD GPUs from RX460 to RX 580 under PCLinuxOS (a .rpm based distribution - not .deb). Over a year ago (mainly through trial and error) I managed to extract a very small subset of version 16.60 of AMDGPU-PRO for Red Hat and install just those OpenCL bits on top of the PCLinuxOS repo version of the open source amdgpu driver and get the whole thing to work. I really didn't understand what I was doing but I did get it all to work. I've done the same with 17.50 from December last year - it was different but I managed to work it out and test it. There was little (if any) change in compute performance. Recently, I did the same with 18.10 which is pretty much the same layout as 17.50. I haven't got around to testing performance yet. I'm part of the "if it ain't broke don't fix it" brigade so most of my GPUs still run the 16.60 stuff :-).

The OP should just grab 18.20 and give it a try. I imagine there must be some standard way to uninstall the previous components first.

---

Cheers,
Gary.

Another user getting 7.9 from Ubuntu default repository.

Thank you all very much for your help! This has been very informative.

After a few hours of head scratching I was able to get BOINC working with an AMD graphics card. First I installed the linux kernel image and header from: https://github.com/M-Bab/linux-kernel-amdgpu-binaries and then the amdgpu-pro graphics stack with openCL support.

Same problem, everytime I install a new version of Debian, boinc loses the GPU(s)!

lspci | grep VGA :

"01:00.0 VGA compatible controller: NVIDIA Corporation GM107 [GeForce GTX 750 Ti] (rev a2)"

boinc-manager log :

"Wed 27 Jun 2018 22:04:33 CEST | | Starting BOINC client version 7.6.33 for x86_64-pc-linux-gnu
Wed 27 Jun 2018 22:04:33 CEST | | log flags: file_xfer, sched_ops, task
Wed 27 Jun 2018 22:04:33 CEST | | Libraries: libcurl/7.52.1 OpenSSL/1.0.2l zlib/1.2.8 libidn2/0.16 libpsl/0.17.0 (+libidn2/0.16) libssh2/1.7.0 nghttp2/1.18.1 librtmp/2.3
Wed 27 Jun 2018 22:04:33 CEST | | Data directory: /var/lib/boinc-client
Wed 27 Jun 2018 22:04:33 CEST | | No usable GPUs found
Wed 27 Jun 2018 22:04:33 CEST | | Host name: tomato
Wed 27 Jun 2018 22:04:33 CEST | | Processor: 12 GenuineIntel Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz [Family 6 Model 158 Stepping 10]
Wed 27 Jun 2018 22:04:33 CEST | | 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 aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx 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 invpcid_single kaiser tpr_shadow vnmi flexpriority ept vpid 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
Wed 27 Jun 2018 22:04:33 CEST | | OS: Linux: 4.9.0-6-amd64 stretch
Wed 27 Jun 2018 22:04:33 CEST | | Memory: 15.60 GB physical, 15.92 GB virtual
Wed 27 Jun 2018 22:04:33 CEST | | Disk: 9.10 GB total, 7.89 GB free"

I can add at least one more NVIDIA GTX750 (two if the power supply will stand them) but (there's always a BUT with Debian) I've got three screens running on the single card, no problems.

Why do all the people who run BOINC with multiple GPUs always have "anonymous accounts"? I've been trying for years to get a completely working machine. I have another machine with the same motherboard, CPU and memory plus two NVIDIA GTX1070 Ti GPUs, but no point in switching it on since the motherboard says it needs Windo$e10, works OK with Windoze7 but Microsoft have disabled the update system!

If I "apt-get install boinc-client-nvidia-cuda" (as root) the cards work and the machine dies! Normally the motherboard, power supply or hard drive burn out! I've got through 3 motherboards, 2 power supplies and 7 2Tb/3Tb hard drives! I can't afford to keep replacing things.

An attempt to contact the author of "boinc-client-nvidia-cuda" Gianfranco Costamagna <[email protected]> I received this reply :

"what boinc-cuda does, is to force the installation of the non-free nvidia drivers.

If such drivers have issues, boinc has no fault at all :)
They are probably binary-only releases, so even Debian has not much to do here, just
reporting to nvidia might help.

BTW you don't need to install boinc-cuda to use cuda stuff, you can also manually
download the driver from nvidia website and start boinc, that will automatically detect it
and use it.

hope this helps,

Gianfranco"

Tried NVIDIA's site, couldn't make any more sense out their offers of help than I could that of Gianfranco, sorry I'm just a user not a software writer!

I need a fairy-godmother/father who understands how these things work please.

---

I'm not a Linux specialist, but this is what I've picked up from message boards over the years:

The Linux kernel and the video driver are tightly bound together. If you update Linux (by itself), that linkage gets broken. The video driver - especially the special bits of the driver we use for computing - stops working.

The result is: you have to (re-) install the video driver after each Linux update, to re-establish the link between them.

Thanks very much for your speedy reply, I've never understood the NVIDIA update system, and someone else said one had to go to debian.org - but I can find nothing there to help either. Result of search :

"Debian Website search

Search for:

Hits per page:
Language:

Syntax help

Contact Raphael Geissert for comments. Search powered by Xapian.
Copyright © 1997-2003,2007,2008,2011,2015 SPI; See license terms. "

---

You might try https://wiki.debian.org/NvidiaGraphicsDrivers?

Thanks Richard

Been through that and installed, identified several things, but still "No usable GPUs found".

The machine is running beautifully and I can't afford to lose all my work as I recently did, (nineteen years of it), I'm getting too old to start yet again so I'll leave well alone, my three screens work. I've wasted nearly four thousand Euros so far this year on equipment that I can't use, I'll just admit defeat and run my 12 CPUs.

Many, many thanks for all your efforts it really is appreciated, perhaps if I can save up for the end of the year I can buy some copies of Windo$e10, much as that goes against the grain.

yours sincerely

Malcolm Beeson

Same problem, everytime I install a new version of Debian, boinc loses the GPU(s)!

lspci | grep VGA :

"01:00.0 VGA compatible controller: NVIDIA Corporation GM107 [GeForce GTX 750 Ti] (rev a2)"

I think we're talking about an upgrade from Debian jessie to stretch here. One thing that comes to mind is that all the fglrx stuff is no longer available in stretch. If your old installation relied on anything of that, I think the result may not be fully functional, depending on how you upgraded. But never mind, you'll probably want to do a fresh install of the nvidia driver anyway. Debian stretch comes with everything required, for the GTX 750 at least.

I can add at least one more NVIDIA GTX750 (two if the power supply will stand them)

If the power supply will stand them? If there's any doubt, don't even try it! Pushing your hardware to the limits, and possibly beyond, is a great way to burn it. And a dying PSU can kill anything in your system. Sounds familiar?

I have another machine with the same motherboard, CPU and memory plus two NVIDIA GTX1070 Ti GPUs, but no point in switching it on since the motherboard says it needs Windo$e10

They'll always tell you you need the latest Windows for anything to work.

works OK with Windoze7

See? You can't give a d*n on what they say. Since you already have the hardware, why not at least try it with Linux if Windows 7 or 10 are no options for you? That stuff is too expensive to just sit idle.

If I "apt-get install boinc-client-nvidia-cuda" (as root) the cards work and the machine dies! Normally the motherboard, power supply or hard drive burn out! I've got through 3 motherboards, 2 power supplies and 7 2Tb/3Tb hard drives! I can't afford to keep replacing things.

I've never seen BOINC kill any hardware. How would it do that? It can't put more stress on the hardware than anything else can, and if your system can't take 100% load it's defective from the start.

I don't really know how you can burn that much hardware, but one thing I could imagine is a not properly dimensioned or not properly connected PSU. There's more to look at than just a nice big total power output. Another thing could be heat. Multi GPU crunchers can generate a lot of heat, and you need to take care of that.

Back to the original topic, running BOINC with Nvidia GPUs on Debian stretch in this case. I added a Nvidia GPU to a headless cruncher only a few days ago, and all I had to install in addition to the already installed plain boinc-client package (which in your case probably is boinc aka boinc-client + boinc-manager) was nvidia-opencl-icd. That pulled in everything required. You'll probably have to do some configuration with your X setup, but computation works just like that.

Don't install any of the boinc-client-whatever packages. They're there for convenience, but if you don't exactly understand what they do they just add to the confusion. And don't install the original Nvidia driver. Sure, first thing everyone tells you is to go to nvidia.com and install the latest binary blob from there, but if your distribution's packaged driver is recent enough, it's much more convenient and less error-prone.

When you install nvidia-opencl-icd, you don't only (obviously) want its dependencies, but also the recommended packages. That way you eventually get nvidia-kernel-dkms, which keeps the Nvidia kernel module up to date. No more need to reinstall the driver after kernel upgrades. But apt-get doesn't subsequently install the recommendations of already installed packages, so you better make sure there's no remnants of old driver installations before you install nvidia-opencl-icd.

I don't think the driver in stretch is recent enough for a 1070 Ti, but the one in stretch-backports should do. There's a possible pitfall however. Just ask if you need to know.

Phew!

nvidia-opencl-icd did the trick, I'll keep an eye on the temperatures for 24hrs and if all is OK, Monday morning I'll add the second GTX750 Ti. I've a 900w power supply so there will be no problem. All being well, I'll dump the W7 running on my other machine with the two GTX1070 Ti GPUs and switch to stretch, the power supply there is a 1200w so no problem there.

Thank you very much for your detailed reply, I understand a little more now.

I'll keep you in the loop, thanks you for your time and help, it is much appreciated.

Malcolm Beeson
P.S. Before I go steaming off into the wild blue yonder, I'll concentrate on the second card for this machine! I've never managed more than one GPU on a Linux machine, have to add "use all gpus" in the config file I think. Thanks again Floyd.

---

To Floyd & Richard
Again my heartfelt thanks. I've been running twelve cores and one GPU all weekend and as I expected there is a heat issue. I've dropped BOINC to 60% (from my normal 75%) and the cores all remain around 70°C which is acceptable for an i7 I believe. Adding the other GPU I feel would be unwise until the summer heat has passed, so now I go into "big pause" mode.
My daily returns have increased considerably with just one GPU so I'm a happy little boinc-er!
Regards
Malcolm

---

Glad you got it working, but I'd like to add a short off topic remark. Maybe you should revise your cooling solution. I see no reason why just a CPU and a single GTX 750 wouldn't be able to run full throttle, even in summer, unless you live in a particularly hot place.

I live in the South of France close by the Mediterranean Sea, we have 30°C shade temperature at the moment. I bought an enormous cooling fan for the CPU (nearly €100), next step was external water cooling. I have another fan in the room running 24/7 and the window is never closed.
"sensors" output:
asus-isa-0000
Adapter: ISA adapter
cpu_fan: 0 RPM

acpitz-virtual-0
Adapter: Virtual device
temp1: +27.8°C (crit = +119.0°C)
temp2: +29.8°C (crit = +119.0°C)

coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +78.0°C (high = +82.0°C, crit = +100.0°C)
Core 0: +75.0°C (high = +82.0°C, crit = +100.0°C)
Core 1: +72.0°C (high = +82.0°C, crit = +100.0°C)
Core 2: +77.0°C (high = +82.0°C, crit = +100.0°C)
Core 3: +77.0°C (high = +82.0°C, crit = +100.0°C)
Core 4: +75.0°C (high = +82.0°C, crit = +100.0°C)
Core 5: +75.0°C (high = +82.0°C, crit = +100.0°C)

---

Due to the heat here in Southern France I've been running BOINC with just one NVIDIA GTX750 Ti and the CPUs at 50%, have now had to reduce the number of hours per day to 16. Did the update (as I always do at least once a day), machine threw a wobbly, said "the NVIDIA driver was not compatible with this version of Linux (9/stretch) and that I would need to reboot, did so, yet again I have a machine that does not boot! Back to same situation, Linux DOES NOT RUN BOINC!.

---