USing a GTX 980 on Linux?

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0
Topic 198585

I removed an Nvidia 750, which was working fine, and installed an Nvidia 980.

The computer works fine except boinc says missing gpu and will not resume gpu work units. Cpu units are working fine.

I have rebooted the computer and tried a boinc client restart.

Any ideas? Running Linux Mint 17, computer is https://einsteinathome.org/host/11454899

Phil

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

USing a GTX 980 on Linux?

Quote:
Any ideas? Running Linux Mint 17

I'm not sure why - but boinc is not detecting the new GPU.

and don't forget

sudo nvidia-modprobe Lots of other reasons, i would suggest starting with the LocutusOfBorg's ppa to get a more up to date boinc, 7.6.22 onwards i recall is better at GPU detection.

hth

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

I'm running the latest stable

I'm running the latest stable version of boinc for linux. That's one of the first things I checked. So far it's got me stumped.

I've tried different drivers, etc.

Phil

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5845
Credit: 109961348573
RAC: 31021070

RE: I'm running the latest

Quote:
I'm running the latest stable version of boinc for linux.


That version is quite old. Berkeley aren't building newer versions for Linux, it seems. Unless you have a specific reason for staying with it, you would be much better off going with what AgentB has suggested.

The reason your GPU is 'missing' is probably that BOINC can't find the OpenCl lib (LibOpenCL.so) wherever it happens to be installed. The solution may be as simple as creating a symbolic link to point to the actual lib from the place where BOINC is trying to find it. As an example, for my distro, the full path to the lib is /usr/lib64/fglrx-current/libOpenCL.so.1. I use BOINC 7.2.42 from Berkeley which wants to find the lib at /usr/lib64/libOpenCL.so. So I just cd to /usr/lib64 and create the link by issuing the following command, as root,

ln -s fglrx-current/libOpenCL.so.1 libOpenCL.so followed by

ldconfig -X The second command adds the new information to the cache so that it can be used immediately without a reboot.

You may find that the more recent BOINCs from the ppa may very well be more intelligent in finding the OpenCL lib so that just upgrading BOINC may lead to everything working once again. I have no experience with Ubuntu or derivatives so can't give you a specific answer.

Cheers,
Gary.

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

@Gary - The boinc I was

@Gary - The boinc I was referring to is the same version you are using. 7.2.42.

I tried your suggestions to no avail, however, mint is having trouble updating so I think there is something about the newer model card my software does not like. I have not done any updates in quite awhile so probably my fault.

If mint is no longer able to update itself it may be time to look elsewhere. I believe you gave me some suggestions sometime in the past, I'll have to go poking around again and see what's out there.

The point of this whole thing, other than raising my RAC, was to see how a 980 would perform on older hardware. We'll see what happens.

BTW, the card in question is an actual Nvidia GTX 980.

Phil

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

RE: I tried your

Quote:
I tried your suggestions to no avail, however, mint is having trouble updating so I think there is something about the newer model card my software does not like. I have not done any updates in quite awhile so probably my fault.

OK so maybe the updating broke the host, not the card swap?

This may not be possible but if you reverse back to the old GPU - does it work?

When you say trouble, do you get an error - if so what is that?

Can you confirm you have installed and run nvidia-modprobe?

How exactly - are you stopping and starting boinc?

Can you
a) post the first 40 lines of the event log
b) he output from nvidia-smi
c) output from xhost

This thread has some minty flavoured examples https://einsteinathome.org/node/197867

boinc was updated around 7.6.22 to be smarter in GPU device and library detection, and importantly better diagnostics.

Christian Beer
Christian Beer
Joined: 9 Feb 05
Posts: 595
Credit: 125735082
RAC: 342140

Also keep in mind that you

Also keep in mind that you need the newest 159 driver to work with the 980. The 140 driver won't work.

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

The card is what broke the

The card is what broke the system. I simply took out a 750, put in the 980, and turned the machine on without changing any software or updating anything.

The machine is completely offline for now. Mint refused to update so I tried installing the latest version and apparently 17.3 has known problems with a 980.

Sorry I can't post the logs. I wiped the system and tried a fresh install. I'm not known for much patience with software. I've found with my knowledge level it's faster to just reinstall. That usually works but this is one of those times it turned around and bit me. :-)

You should all know I tried all of your suggestions so far except for posting logs and installing the old card.

I know people are using 980s under linux for crunching, so I just need to find out what they are using. Mint has treated me well but apparently is broken for the newer cards.

Could a moderator perhaps change the name of the thread to "Using a GTX 980 with Linux" ? or some such name? Once I get this solved it will make more sense to people having similar problems.

Phil

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

RE: Could a moderator

Quote:
Could a moderator perhaps change the name of the thread to "Using a GTX 980 with Linux" ? or some such name? Once I get this solved it will make more sense to people having similar problems.


You can do that yourself as the OP of the thread, just post a new message to the thread, don't reply or quote, then edit that message and you should be able to change the title of the thread.

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

RE: The machine is

Quote:

The machine is completely offline for now. Mint refused to update so I tried installing the latest version and apparently 17.3 has known problems with a 980.


ah ok, can you post a link - Mint and 980 is likely to come up in the future.

Quote:

Sorry I can't post the logs. I wiped the system and tried a fresh install. I'm not known for much patience with software. I've found with my knowledge level it's faster to just reinstall. That usually works but this is one of those times it turned around and bit me. :-)

NP - I'm a firm believer in reinstalling, it's really hard to weed out previous artefacts and apps from previous OS, and it keeps your OS new. The leading edge on linux, is sometimes a little wobbly.

Quote:
You should all know I tried all of your suggestions so far except for posting logs and installing the old card.

thanks.

Quote:

I know people are using 980s under linux for crunching, so I just need to find out what they are using. Mint has treated me well but apparently is broken for the newer cards.

Could a moderator perhaps change the name of the thread to "Using a GTX 980 with Linux" ? or some such name? Once I get this solved it will make more sense to people having similar problems.

I've often wondered how you edit a title, maybe put (tried Mint 17.3) in the subject too.

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

Once I get this sorted out

Once I get this sorted out I'll make a new thread with instructions on getting it to work. I'm trying out various versions and even though I have made it to a desktop, so far nothing is playing very nice.

Can't remember which link for the 980 mint problem. I've been all over the place.

Try searching "linux mint gtx 980" that should find several links with problems.

Phil

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.