Advanced search

Message boards : Number crunching : Ubuntu 12.10, nVidia drivers, libcuda1 and BOINC

Author Message
Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28105 - Posted: 22 Jan 2013 | 1:40:31 UTC
Last modified: 22 Jan 2013 | 1:41:28 UTC

I upgraded from Kubuntu 11.10 which worked very well for crunching GPUgrid to Ubuntu 12.10 with Unity desktop which is proving difficult. Does anybody know how to set this up?

I installed BOINC 7.0.27 from Ubuntu 12.10 repos and tried to install libcuda1 (the nVidia driver package) from repos as well. The package installer says the driver package is referenced in one other package but does not name that package which I suspect is boinc-nvidia-cuda. Attempts to install boinc-nvidia-cuda result in the same error as for libcuda1.... no candidate package, the package seems to have been in repo at one time but is now missing.

Then I dug up a 3rd party libcuda1 .deb package and installed that. After that BOINC recognized the GPU on 3 out of 4 BOINC starts but did not recognize it on one start. That suggests there is a problem. And there is definitely a problem because even when BOINC recognizes the driver it requests tasks from GPUgrid but the server never sends any.

The nvidia-settings app works which suggests the driver is installed and at least semi-working.

Now I have removed the 3rd party libcuda1 driver and have downloaded the recommended driver from nVidia... 310.32. It's one of those .run files which you must run in a CLI session with no X session running. The problem is I cannot boot to a CLI sesion. When I reboot there is no menu that allows choice between GUI and CLI session. Another way that has always worked with every Linux distro I've used to date is the init command... init 3 will close a running X session and drop to CLI. That doesn't work with Ubuntu 12.10. The third way, which is to edit /etc/inittab and specify runlevel 3 as the default runlevel at boottime, doesn't work either for there is no /etc/inittab in stock Ubuntu 12.10. The reason appears to be that /etc/inittab is old Sys V protocol whereas they are migrating to Upstart. Man init tells me I can create an /etc/inittab and put an id:3:initdefault line in it which will cause init to boot torunlevel 3 but that doesn't work either.

That's the point I'm stuck at. If anybody is crunching GPUgrid with Ubuntu 12.10 and Unity then I'd appreciate hearing how to do it. I've Googled around for hints/solutions and have turned up nothing, it isn't addressed in any of the forums I checked (or I missed it).

I'll press on and get it eventually then post the solution here. Stay tuned if interested. I'm hoping they'll find out what's wrong with libcuda1 , fix it and put it back in repos as that would be the easiest. If not I will find a way to boot runlevel 3 or halt an existing X session and install nVidia's driver the old fashioned way.

Profile AdamYusko
Send message
Joined: 29 Jun 12
Posts: 26
Credit: 21,540,800
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwat
Message 28106 - Posted: 22 Jan 2013 | 3:00:26 UTC - in response to Message 28105.

Possibly not the solution you are looking for, but if you are dead-set on Ubunutu, try the Long Term Support offering, that should be 12.04. I just set up another system using 12.04 and it was set up like a breeze.
____________

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28107 - Posted: 22 Jan 2013 | 4:40:17 UTC - in response to Message 28106.
Last modified: 22 Jan 2013 | 5:00:32 UTC

It's mostly stubborness that makes me want to make 12.10 work. In reality 12.04 LTS is more than adequate. Did you use the Berkeley installer or did you install BOINC from repos? If from repos then which BOINC version does it provide?

Edit added:

Ach! It never fails!!! You try, it does not work, you beg for help and let everybody see you are not the guru you pretend to be and then suddenly it starts to work. How does it know??!! And why does it pick on me? ;-)

Edit added 2:

Now there are 2 lines in which BOINC states what it recognizes regarding GPU, before there was only the blue line, now there is the red line as well.

Mon 21 Jan 2013 04:33:09 PM MST | | Starting BOINC client version 7.0.27 for x86_64-pc-linux-gnu
Mon 21 Jan 2013 04:33:09 PM MST | | log flags: file_xfer, sched_ops, task
Mon 21 Jan 2013 04:33:09 PM MST | | Libraries: libcurl/7.27.0 OpenSSL/1.0.1c zlib/1.2.7 libidn/1.25 librtmp/2.3
Mon 21 Jan 2013 04:33:09 PM MST | | Data directory: /var/lib/boinc-client
Mon 21 Jan 2013 04:33:09 PM MST | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
Mon 21 Jan 2013 04:33:09 PM MST | | Processor: 8.00 MB cache
Mon 21 Jan 2013 04:33:09 PM MST | | 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 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
Mon 21 Jan 2013 04:33:09 PM MST | | OS: Linux: 3.5.0-22-generic
Mon 21 Jan 2013 04:33:09 PM MST | | Memory: 23.53 GB physical, 23.98 GB virtual
Mon 21 Jan 2013 04:33:09 PM MST | | Disk: 663.98 GB total, 625.95 GB free
Mon 21 Jan 2013 04:33:09 PM MST | | Local time is UTC -7 hours
Mon 21 Jan 2013 04:33:09 PM MST | | NVIDIA GPU 0: GeForce GTX 570 (driver version unknown, CUDA version 5.0, compute capability 2.0, 134214912MB, 134214839MB available, 1425 GFLOPS peak)
Mon 21 Jan 2013 04:33:09 PM MST | | OpenCL: NVIDIA GPU 0: GeForce GTX 570 (driver version 304.51, device version OpenCL 1.1 CUDA, 1280MB, 134214839MB available)
Mon 21 Jan 2013 04:33:09 PM MST | | Config: GUI RPC allowed from:
Mon 21 Jan 2013 04:33:09 PM MST | | A new version of BOINC is available. <a href=http://boinc.berkeley.edu/download.php>Download it.</a>
Mon 21 Jan 2013 04:33:09 PM MST | GPUGRID | URL http://www.gpugrid.net/; Computer ID 144260; resource share 20
Mon 21 Jan 2013 04:33:09 PM MST | | General prefs: from http://sat.isa.ru/pdsat/ (last modified 05-Jan-2013 18:10:28)
Mon 21 Jan 2013 04:33:09 PM MST | | Host location: none
Mon 21 Jan 2013 04:33:09 PM MST | | General prefs: using your defaults
Mon 21 Jan 2013 04:33:09 PM MST | | Reading preferences override file
Mon 21 Jan 2013 04:33:09 PM MST | | Preferences:
Mon 21 Jan 2013 04:33:09 PM MST | | max memory usage when active: 24092.30MB
Mon 21 Jan 2013 04:33:09 PM MST | | max memory usage when idle: 24092.30MB
Mon 21 Jan 2013 04:33:09 PM MST | | max disk usage: 20.00GB
Mon 21 Jan 2013 04:33:09 PM MST | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
Mon 21 Jan 2013 04:33:09 PM MST | | Not using a proxy
Mon 21 Jan 2013 04:33:10 PM MST | GPUGRID | Restarting task I4R95-NATHAN_RPS1120528-113-166-RND9252_0 using acemdlong version 617 (cuda42) in slot 0

RaymondFO*
Send message
Joined: 22 Nov 12
Posts: 72
Credit: 14,040,706,346
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28108 - Posted: 22 Jan 2013 | 5:11:25 UTC - in response to Message 28106.
Last modified: 22 Jan 2013 | 5:26:16 UTC

Problem has been solved. Posting has been removed.

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28121 - Posted: 22 Jan 2013 | 17:11:17 UTC
Last modified: 22 Jan 2013 | 17:12:49 UTC

Hello: BOINC 7.0.28 works perfectly in Ubuntu 12.10.

It is best to install it directly using "boinc_7.0.28_x86_64-pc-linux-gnu.sh" we must bear in mind that the following libraries are missing:

libcurl3

libwxgtk2.8-0 * this is most often missing, install from Synaptic easy.

libbase


NVIDIA drivers, Ubuntu 12.10 needs to be installed first "linux-headers-generic".

In the terminal: "sudo apt-get install linux-headers-generic"

Once downloaded you can run the installation with Additional Drivers in Software Sources, no problem . Greetings.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28149 - Posted: 23 Jan 2013 | 7:02:06 UTC - in response to Message 28121.
Last modified: 23 Jan 2013 | 7:04:11 UTC

Thank you Carlesa25! Especially for the packages for the missing libs.

I still have a problem with the GPU not being recognized sometimes when the OS boots. I think it's because the BOINC daemon starts before X server is fully up and running. The rig boots very quickly because /boot is on a hybrid drive. I think I will try putting a sleep in the BOINC daemon init script to delay it until after X server is fully up and initialised or perhaps adjust the order of the scripts in init.d.

If that does not work I will follow your suggestion to install from the Berkeley installer, thank you.

A question...

Do you have manual fan speed control enabled? How? Before my upgrade to Ubuntu 12.10 I used the "Coolbits 5" and another option in /etc/xorg.conf to enable manual fan speed adjustment. Now on 12.10 there is no /etc/xorg.conf and I cannot find a replacement for it or any way to enable manual fan control. My GTX 570 runs at 60C now but I will not be able to maintain that in the summer unless I can boost the fan speed.

If all else fails I'll just rewire the fan to my own PWM control and connect the GPU's fan leads to some dummy fan to make it think it still has a functioning fan but a software solution is less trouble.

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28155 - Posted: 23 Jan 2013 | 14:45:00 UTC - in response to Message 28149.

Hello: If we can control the fan with a card as well with two (harder) in Ubuntu 12.10.

For a single card can use "Coolbits 4" in "etc/X11/xorg.conf" for more than one card I have posted a tutorial on how to turn on more than one Coolbits Nvidia GPU in Ubuntu 12.10.

To save the configuration of nvidia-settings and reload the restart must uncheck the "Include X Displays name .... '

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28156 - Posted: 23 Jan 2013 | 16:06:55 UTC - in response to Message 28155.

But I don't have /etc/X11/xorg.conf in Ubuntu 12.10. In Kubuntu 11.10 there was an /etc/X11/xorg.conf and I added "Coobits 5" to it but on 12.10 there is no /etc/X11/xorg.conf. Do I have to create it myself? If I must create it myself then please can you post your xorg.conf as an example?

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28157 - Posted: 23 Jan 2013 | 16:29:20 UTC - in response to Message 28156.

Hello: The first is to confirm if you have "nvidia-settings" installed, if so just go to "X Server Display Configuration + Save to X configuration file" generated xorg.conf file in etc/X11.

Then you can edit and modify adding = Option Coolbits "4".

The link below is the tutorial I posted for fan control when more than one Nvidia GPU installed.

http://www.canalboinc.com/modules/newbb/viewtopic.php?topic_id=285965&forum=150

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28158 - Posted: 23 Jan 2013 | 16:40:52 UTC - in response to Message 28157.

Doh! Thanks, now I understand :-)

Thanks for the link to the tutorial too. I will add all the information you have provided to the official BOINC wiki, very helpful.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28188 - Posted: 24 Jan 2013 | 12:00:46 UTC

There we go! My GTX 570 is running at 49C with stock onboard cooling, fan at 85%, graphics clock at 742 MHz, memory clock at 1900 MHz and processor clock at 1484 MHz. Actually that's up 4 degrees from yesterday, was 45C then.

Now it's ready for some OC.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28287 - Posted: 29 Jan 2013 | 22:31:08 UTC

It ran perfect for several days then I tried to upgrade to the 313.8 driver. That failed and after much playing around I discovered it's not the 313.8 driver it's adding the Option "Coolbits" "4" line to xorg.conf that causes the failure.

Now the first method I used to install the driver no longer works because the libcuda1 package seems to have changed. It was absent from official Ubuntu repos and the one I used is a third party package and is classified by the provider as experimental. Also, the 310.x driver installation method mentioned by Carlesa25 in this thread works but suffers from the same problem with the Coolbits line as does the 313.x driver.

Anyway, I have it working (crunching) again though I don't have the manual fan control I want which is not a big deal. I have a hunch these drivers are being updated on a daily or perhaps weekly basis so if I wait a while the issues I seem to be encountering with the Coolbits option might be fixed.

For the manual fan control I think what I will do is modify the card and put the fan onto my own speed controller or just wire it straight to +5 volts (or +12 if that's what it needs) and remove the need for the Coolbits 4 option entirely. That would eliminate another problem as well which is that one cannot use Coolbits 4 and Coolbits 1 simultaneously. I also want to be able to use Coolbits 1 to allow overclocking. I'll report on success or (heaven forbid) failure of that attempt in a different thread.

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28293 - Posted: 30 Jan 2013 | 0:04:26 UTC - in response to Message 28287.

Hello: I do not understand the problem you have with the option Coolbits 4 me is working perfectly with the 313.18 driver in Ubuntu 10.12-64bit as with the above.

I have installed GTS450 and GTX590 fan controlling each.

Could you show the file /etc/X11/xorg.conf to see how it is configured.

There is an old problem that can not save settings Nvidia-Setings if we checked the box "Include X Display Names ..... "When reinstalling a new driver is recommended to run" Save Current Configuration "to generate a new configuration file before restarting ... failure could be that ...? Greetings.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28296 - Posted: 30 Jan 2013 | 0:59:56 UTC - in response to Message 28293.

Yes, that might be the problem, I don't know. I discovered that if I simply comment out the Coolbits 4 line with a # then it will work again so that tells me the xorg.conf was saved correctly and it is only the presence of the Coolbits 4 which causes it to not boot.I don't understand it either because the Coolbits 4 worked perfectly for me too when I first installed the driver using the libcuda1 package I found at the third party source. Now, after reinstalling Ubuntu 12.10, Coolbits does not work and I don't understand it either. I suspect there have been recent changes that have broken the driver with respect to Coolbits. Maybe it cannot parse the file correctly and it crashes when it reads the Coolbits line, I don't know, like I said:

1) I install using your method
2) everything works, it boots, BOINC sees the GPU
3) I add the Options "Coolbits" "4" in the Device section, I double check to make sure no typos,
4) I try to reboot, but it won't boot, I get the "Running at low resolution..." message
5) I go to command prompt and edit xorg.conf and comment out the Coolbits line
6) then it boots and works again, BOINC sees the GPU
7) then i edit xorg.conf, check everything, no typos, everything is correct, I remove the # to activate the Coolbits line
8) I rebbot but again it will not boot and I get the "Running at low resolution..." message
9) then I edit xorg.conf and comment out the Coolbits line again and then it boots and everything works

Anyway, I am no longer interested in making Coolbits 4 work. I will hack the card and connect the fan to my own controller so that I will have manual control even without Coolbits 4. And that way I will be able to put the fan up to 100% speed instead of this stupid 85% limit imposed by nVidia. No, I don't care if it is noisy. I don't want Coolbits 4 anymore because if I use it then I cannot use Coolbits 1. I need Coolbits 1 to allow OC so goodbye Coolbits 4.

Post to thread

Message boards : Number crunching : Ubuntu 12.10, nVidia drivers, libcuda1 and BOINC

//