CUDA device not recognised by BOINC ? try this....

[veebee]

After mucking around installing/ unistalling drivers, programs to install drivers etc etc, I just could not get my 9600 and 9500 recognised in Boinc to allow me to utilise them on GPUgrid.

Mike047 sent me these instructions and had allowed me to post them for others , who may be in the predicament I was, to use.

I have added in red, anything I "did differently"...

QUOTE from Mike047:

Install Ubuntu 8.04.    (I didn't go for a re-install.... just went to the next step)

Download and install all updates. [reboot required]

Go into user group[sys/admin/usr /grp]
Give root a password and authorize it for everything.

Download and install;
IA32-libs
build-essential       (I already had these installed)

Find and "remove" all Nvidia restricted drivers[I think this is key]   (I just used "Synaptic", searched for "NVIDIA" and selecyed anything that WAS installed, for complete removal)

Download appropriate Nvidia driver[ I use either 177.82 or180.22]    (I used the latest, 180.52)
place in a folder in home folder.                                                          (I just saved it to the Desktop)

Download appropriate Boinc client[I use 6.4.5]
place in a folder in home folder.                         I used my current BOINC installation. I just shut the client down with "sudo /etc/init.d/ boinc-client stop")

press ctrl+alt-F1
This will take you to a "real" terminal

Login as root, then the newly created password

next;
sudo /etc/init.d/gdm stop
this stops your desktop and will allow installation of Nvidia drivers.

next;
cd /home/your user name/Nvidia folder you created       ( I just "cd'd" to the Desktop - remember the "capital "D"in Desktop ... 

next;
sh NVIDIA-Linux-x86_64-driver number-pkg2.run            (insert the version number of the driver you have downloaded)
accept terms and answer ok or yes to everything.

you will be returned to a terminal, enter;
sudo /etc/init.d/gdm start

You should see the Nvidia splash screen as it returns you to your login.  If you don't see the splash screen...everything is probably stuffed.     (I didn't see the splash screen.. maybe it was too quick for me, though I have seen it before and it was up for a cpl seconds...so effectively, I imagine the splash screen DIDN'T appear in my install)

Log back in.             ( Here is where I noticed the first "difference"... my screen resolution, which had been stuck on 1600 x 1200 (instead of 1680 x 1050) now looked "right" - and is  "right")

Go to the folder with your Boinc client and run it. it will unfold to a folder"boinc'.                               ( Again, I didn't use a new install of "BOINC", here, I simply restarted from a Konsole, with "sudo /etc/init.d/boinc-client start")
Click boinc manager in this folder and Boinc will start.  It should recognize you cuda capable card.     ( I went to the "messages" tab in Boinc manager and Lo and Behold... "CUDA device found " !!!!! )

The plus to running boinc out of a dedicated folder in home is the ease of changing it.  The drawback is that you must start it there or from a folder link on your desktop[ this is the way I do it].  You can go to "sessions" [sys/pref/sessions] and make boinc-client and boinc-manager start on boot from there.  I have had good luck and bad luck with that method.   I find it easier to just start if from a desktop link.


END QUOTE FROM Mike047

So there you have it....considering I didn't have to re-install the OS, or BOINC itself, it was a fairly straightforward, easy fix.

Hopefully it will help anyone else who may find themselves stuck with their Graphics card not being recognised by BOINC.


A Big thanks once again, to Mike047, for this help. And I believe anyone who uses it should happily give Mike a "+1" Karma....

[Rob]

As mentioned elsewhere the hard drive in my Q9550 kicked the bucket last week. Luckily I had a spare so installed Kubuntu 9.04 on it & got everything running & low & behold No Cuda devices found. So did everything I did last time & eventually got it working.
Everything was fine for a couple of days until I had another power outage. On restart I once again had No cuda devices found. After fiddleing around on & off for a few days without success I decided to bite the bullet & try Veebees solution. It didn't go well.

Go into user group[sys/admin/usr /grp]
Give root a password and authorize it for everything.

What does this mean? I went into user manager & added everything to the root group but doesn't root already have a password when you install?(your login)

I used my current BOINC installation. I just shut the client down with "sudo /etc/init.d/ boinc-client stop")

That didn't work tried using "boinc" instead of "boinc-client" that didn't work either. Must be a different command in Kubuntu ? Anyway I ended up shutting it down with Boinc manager.

sudo /etc/init.d/gdm stop

Changed "gdm" to "kdm" & it worked fine.

sh NVIDIA-Linux-x86_64-driver number-pkg2.run            (insert the version number of the driver you have downloaded)
accept terms and answer ok or yes to everything.

Had to run it as Sudo before it would work. Then it kept hanging on Downloading the Kernel so I had to skip that & make it compile it's own.
When I finally fired it back up, still no Cuda devices found :hbang:. Tried a reboot just in case. Still no good :cry2:

[veebee]

Rob, yeah, "root" probably already has a password, but that's fine. you need to go into "users and groups " ("system" menu > "Users and groups" or similar in Kubuntu), the "unlock" it so you can 'edit " the users... go to edit "root", and you will see where yu can authorise root (and other users) for various processes etc. Authorise EVERYTHING for rott.


it should be the same commands.. (has to be "boinc-client") unless you have Boinc installed somewhere else on your machine. Shutting it down by the menu ("Advanced" > "shut down connected client") is fine - same thing.

Shutting down kdm was no prob.

When you go to the "real terminal", you must log in as "root" (and enter password) ... that will allow you to run it, coz "root" is authorised to run everything.

as for why it wouldn't/ couldn't download kernel I dont know.. have you got the right/ latest driver? are you putting the right driver number in the "sh NVIDIA... etc?"

I still have to do this everytime I need reboot a machine, except one for some reason.... but to be honest, one of them has been "uptime" for over a cl weeks now, so it may well "keep the driver installed" too.. dunno.


So, I cannot see why it wouldn't work (especially considering you got to ythe point of the driver installation itself..), all I know is it works evrytime I use this method.. without fail.

[Wang Solutions]

That command for stopping boinc only works if you have done the synaptic install. If you have done it manually or in another location, it will not work.

[Rob]

I did the synaptic install but have copied the latest boinc, boinccmd, & boincmgr to /usr/bin.

[veebee]

That command for stopping boinc only works if you have done the synaptic install. If you have done it manually or in another location, it will not work.

that's basically what I said to Rob, but differently...

it should be the same commands.. (has to be "boinc-client") unless you have Boinc installed somewhere else on your machine. Shutting it down by the menu ("Advanced" > "shut down connected client") is fine ...

I gave up "moving things around" etc, as I found that I had to start remembering to do all sorts of extra things at update times, and manually restarting things etc.
I now leave everything where Synaptic installs it.
In regards to upgrading BOINC "manually".. there was a post by (i think) CP about how to do a manual upgrade of boinc (higher than the repository version) and to make sure everything went where it "should go"......  I used it and it worked, but the version of BOIONC i went to had bugs with CUDA app etc, so I went back to 6..4.5.

[Rob]

The only thing I can think that it would be is the downloading the kernel thing as everything else eventually worked. How long should downloading the kernel take? The longest I gave it was 5-10 minutes & there was no activity on the screen or on my network switch(no flickering lights).
I might try again on the weekend if I get the time. Only problem is how do I uninstall the drivers when they've been installed this way as they don't show in Synaptic?

[veebee]

there's about 4 or so "progress bars" occur during the whole processs... but I didn't know that anything is actually being downloaded at anytime... I already have the driver file(s) downloaded  and install from them.

each of these progress bars get to 100% pretty quick, though a cpl DO 'hang " for a second... but none of them last even close to one minute.

As for uninstalling the drivers.. why would you ? what you DO want, is the latest drivers, direct from the "source" , installed on your machine/s.

The whole process from shutting down BOINC, exiting to command prompt, shutting down gdm (kdm), doing the install, the program "building" the kernel etc, and restarting the GDM, takes about 5 minutes..... actually less now, as I have just realised that all the "commands" are "saved"... I didn't need to remember/ retype them all the time like I was..!

[Rob]

but I didn't know that anything is actually being downloaded at anytime... I already have the driver file(s) downloaded  and install from them.

That's what I would have thought, but none the less it asks if I want to download some kernel (can't remember exactly what it was) if I say yes it hangs, & if I say no it compiles it's own which takes maybe a minute at most.

As for uninstalling the drivers.. why would you ? what you DO want, is the latest drivers, direct from the "source" , installed on your machine/s.

Isn't it supposed to be best to uninstall old drivers, or in this case possibly corrupted current drivers, before installing new ones?

[veebee]

the first time I did this system, I went to synaptic and "completely removed" anything to do with nvidia.

since then, I just repeat this procedure whe needed.... which is only after a reboot is required/ happens for some reason.

One of the quads shows th e nvidia splash screen on bootup, and hasn't needed to have the drivers reinstalled for ages, but one of the others requires it every time it is rebooted, and the i7 requires it every other time..

[Rob]

Just tried again with the same result. When I try to shut boinc down with sudo /etc/init.d/ boinc-client stop it says sudo /etc/init.d/ command not found. Yet strangely when I use the sudo /etc/init.d/ boinc-client start command the client will start, although it is a blank slate with no projects. I have to restart the computer so boinc starts automatically for it to find my current work & projects.
Took notice of the kernel thing this time. It's the first thing after you accept the terms. It says no kernel interface was found to match the kernel do you want to download the interface from the NVIDIA ftp site.
The splash screen definitely shows up on restart and Xserver settings shows 180.51 driver installed.
As a last resort I suppose I could uninstall boinc through Synaptic & re-install. Hopefully I shouldn't lose my work. I'll leave that until tomorrow though so if I stuff up I have more time to fix things  ;D

[Mike Mitchell]

Isn't it supposed to be best to uninstall old drivers, or in this case possibly corrupted current drivers, before installing new ones?

That's more of a windows thing. When anything is installed it should overwrite any existing file of the same name, why MS can't manage that is anyone's guess.