Linux GPU setup

Message boards : Number crunching : Linux GPU setup
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
HankB

Send message
Joined: 1 Dec 04
Posts: 12
Credit: 6,115,914
RAC: 1
United States
Message 1315753 - Posted: 15 Dec 2012, 21:44:48 UTC

I had this running on my GTX 460 using a custom GPU app that (I think) I found at Lunatics. This was before I upgraded from Ubuntu 10.04 LTS to Mint 13 (Maya). I copied over the files in /var/lib/boinc-client/ nd hoped it would just work, but unfortunately it did not.

At present I have the following nvidia drivers installed:

hbarta@olive ~ $ dpkg -l|grep 310
ii  nvidia-experimental-310                310.14-0ubuntu0.1                       Experimental NVIDIA binary Xorg driver, kernel module and VDPAU library
ii  nvidia-settings-experimental-310       310.14-0ubuntu0.1                       Tool for configuring the NVIDIA graphics driver


On startup, boinc reports:
Sat 15 Dec 2012 03:16:14 PM CST |  | No usable GPUs found


What else am I missing? I seem to recall that previously I needed to download from nvidia the dev kit and install that. Is that still required or is ther some other setting I have overlooked.

thanks,
hank

ID: 1315753 · Report as offensive
HankB

Send message
Joined: 1 Dec 04
Posts: 12
Credit: 6,115,914
RAC: 1
United States
Message 1315757 - Posted: 15 Dec 2012, 22:01:16 UTC - in response to Message 1315753.  

I found this guide: https://help.ubuntu.com/community/Cuda

It describes installing Cuda 4.2 whereas nvidia lists 5.0 as the current version.

Which should I install?

thanks,
hank
ID: 1315757 · Report as offensive
Jeroen
Volunteer tester

Send message
Joined: 28 Mar 01
Posts: 13
Credit: 8,155,940
RAC: 0
United States
Message 1315894 - Posted: 16 Dec 2012, 6:18:52 UTC - in response to Message 1315753.  

In order for BOINC to detect NVIDIA GPUs in Linux, two things are necessary. One is libcuda.so and the other is the kernel module nvidia.ko. If the distribution is 64-bit, this library could be in either /usr/lib64 or /usr/lib. If you are trying to run 32-bit BOINC in 64-bit Linux, then you will also need the 32-bit NVIDIA libraries including libcuda.so installed which is a question that the NVIDIA driver installer asks. The 32-bit libraries may be in /usr/lib or /usr/lib32 depending on distribution.

To check if nvidia.ko is loaded:

/sbin/lsmod | grep nvidia

The Seti application also depends on two NVIDIA libraries including libcudart.so and libcufft.so. These should come with the application but are otherwise available with the NVIDIA CUDA toolkit. When in doubt, you can run this command on the application to see if all necessary libraries are in place:

ldd setiathome_x41g_x86_64-pc-linux-gnu_cuda32 (replace with the name of the application that you are running)
ID: 1315894 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1315903 - Posted: 16 Dec 2012, 7:15:54 UTC - in response to Message 1315894.  
Last modified: 16 Dec 2012, 7:44:02 UTC

In order for BOINC to detect NVIDIA GPUs in Linux, two things are necessary. One is libcuda.so and the other is the kernel module nvidia.ko. If the distribution is 64-bit, this library could be in either /usr/lib64 or /usr/lib. If you are trying to run 32-bit BOINC in 64-bit Linux, then you will also need the 32-bit NVIDIA libraries including libcuda.so installed which is a question that the NVIDIA driver installer asks. The 32-bit libraries may be in /usr/lib or /usr/lib32 depending on distribution.

To check if nvidia.ko is loaded:

/sbin/lsmod | grep nvidia

The Seti application also depends on two NVIDIA libraries including libcudart.so and libcufft.so. These should come with the application but are otherwise available with the NVIDIA CUDA toolkit. When in doubt, you can run this command on the application to see if all necessary libraries are in place:

ldd setiathome_x41g_x86_64-pc-linux-gnu_cuda32 (replace with the name of the application that you are running)

Nice info, I'll be looking at this later when I try to get my NV8800 working in 64bit Ubuntu. I have a couple of related questions. Could you explain how to install the BOINC 7.0.28 download SETI suggests you download and install after installing 7.0.27? I've not been able to install the downloaded .28 version and have it work without also having .27 'installed'. Assume that you've installed .27 in it's default location and you want to install .28 in your Home Folder. That's about where I am at present. Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version. I've been looking at these threads, Can't get GPU to process workunits on Linux machine and Ubuntu 12.04 and GPU NVIDIA while trying to set up my newest SETI machine.
Thanks.
ID: 1315903 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1315962 - Posted: 16 Dec 2012, 11:34:26 UTC - in response to Message 1315903.  
Last modified: 16 Dec 2012, 11:44:58 UTC

Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version.

There are SSE2, SSE3 and SSSE3 LX64 packages in the downloads section of Crunchers Anonymous.

Claggy
ID: 1315962 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1315982 - Posted: 16 Dec 2012, 12:29:50 UTC - in response to Message 1315962.  

Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version.

There are SSE2, SSE3 and SSSE3 LX64 packages in the downloads section of Crunchers Anonymous.

Claggy

Those appear to be just for AstroPulse tasks, not Multibeam. I had looked through there earlier as well as the old links in those threads. I also did a Google...
It almost seems like someone made the Multibeam version go away, similar to the Lunatics installer. Although you can still find the Lunatics installer if you look hard enough and don't let the Chinese webpages bother you :-)

The 32bit version is working for now. There is still a noticeable improvement over the stock SETI Multibeam App, even though the Anonymous App is from 2008.
ID: 1315982 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1315992 - Posted: 16 Dec 2012, 13:05:33 UTC - in response to Message 1315982.  

Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version.

There are SSE2, SSE3 and SSSE3 LX64 packages in the downloads section of Crunchers Anonymous.

Claggy

Those appear to be just for AstroPulse tasks, not Multibeam. I had looked through there earlier as well as the old links in those threads. I also did a Google...
It almost seems like someone made the Multibeam version go away, similar to the Lunatics installer. Although you can still find the Lunatics installer if you look hard enough and don't let the Chinese webpages bother you :-)

The 32bit version is working for now. There is still a noticeable improvement over the stock SETI Multibeam App, even though the Anonymous App is from 2008.

The PACKAGES contain both MB and AP, note the filename: LX64_AKV8_AP6_AVX_SSSE3.zip

Claggy
ID: 1315992 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1316002 - Posted: 16 Dec 2012, 13:42:11 UTC - in response to Message 1315992.  

Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version.

There are SSE2, SSE3 and SSSE3 LX64 packages in the downloads section of Crunchers Anonymous.

Claggy

Those appear to be just for AstroPulse tasks, not Multibeam. I had looked through there earlier as well as the old links in those threads. I also did a Google...
It almost seems like someone made the Multibeam version go away, similar to the Lunatics installer. Although you can still find the Lunatics installer if you look hard enough and don't let the Chinese webpages bother you :-)

The 32bit version is working for now. There is still a noticeable improvement over the stock SETI Multibeam App, even though the Anonymous App is from 2008.

The PACKAGES contain both MB and AP, note the filename: LX64_AKV8_AP6_AVX_SSSE3.zip

Claggy

I found it. I did look all through there, just didn't find that one page. It's here for anyone else looking: Categories You can see the results here once it's updated All SETI@home Enhanced tasks The ones with the 32bit SSSE3 App came in at 3:18, or around 11,890 verses 17,xxx. I'm being blasted with the longer vlars on all my machines at present. Once I place the thing in a case, I'll install the heavier 8800.
ID: 1316002 · Report as offensive
Jeroen
Volunteer tester

Send message
Joined: 28 Mar 01
Posts: 13
Credit: 8,155,940
RAC: 0
United States
Message 1316107 - Posted: 16 Dec 2012, 17:18:08 UTC - in response to Message 1315903.  
Last modified: 16 Dec 2012, 17:19:56 UTC


Nice info, I'll be looking at this later when I try to get my NV8800 working in 64bit Ubuntu. I have a couple of related questions. Could you explain how to install the BOINC 7.0.28 download SETI suggests you download and install after installing 7.0.27? I've not been able to install the downloaded .28 version and have it work without also having .27 'installed'. Assume that you've installed .27 in it's default location and you want to install .28 in your Home Folder. That's about where I am at present. Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version. I've been looking at these threads, Can't get GPU to process workunits on Linux machine and Ubuntu 12.04 and GPU NVIDIA while trying to set up my newest SETI machine.
Thanks.


This BOINC install should be fairly simple. You can grab any version of BOINC which will be an executable file. Run that file, and the BOINC files automatically extra to a directory called BOINC. Here is a breakdown on how to do this in a shell:

wget http://boinc.berkeley.edu/dl/boinc_7.0.28_x86_64-pc-linux-gnu.sh
chmod 755 boinc_7.0.28_x86_64-pc-linux-gnu.sh
./boinc_7.0.28_x86_64-pc-linux-gnu.sh
cd BOINC
./boinc --allow_remote_gui_rpc (Start the BOINC client)

In the BOINC directory, there is a file called gui_rpc_auth.cfg. This has the key so you can remote connect to the Linux BOINC client via BOINC Manager.

Attach to the Seti project and give BOINC a minute to create the project directory. Then stop BOINC. Extract the Seti archive with the app_info.xml file to the project directory at BOINC/projects/setiathome.berkeley.edu. Start the BOINC client again.
ID: 1316107 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1316154 - Posted: 16 Dec 2012, 18:50:22 UTC - in response to Message 1316107.  


Nice info, I'll be looking at this later when I try to get my NV8800 working in 64bit Ubuntu. I have a couple of related questions. Could you explain how to install the BOINC 7.0.28 download SETI suggests you download and install after installing 7.0.27? I've not been able to install the downloaded .28 version and have it work without also having .27 'installed'. Assume that you've installed .27 in it's default location and you want to install .28 in your Home Folder. That's about where I am at present. Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version. I've been looking at these threads, Can't get GPU to process workunits on Linux machine and Ubuntu 12.04 and GPU NVIDIA while trying to set up my newest SETI machine.
Thanks.


This BOINC install should be fairly simple. You can grab any version of BOINC which will be an executable file. Run that file, and the BOINC files automatically extra to a directory called BOINC. Here is a breakdown on how to do this in a shell:

wget http://boinc.berkeley.edu/dl/boinc_7.0.28_x86_64-pc-linux-gnu.sh
chmod 755 boinc_7.0.28_x86_64-pc-linux-gnu.sh
./boinc_7.0.28_x86_64-pc-linux-gnu.sh
cd BOINC
./boinc --allow_remote_gui_rpc (Start the BOINC client)

In the BOINC directory, there is a file called gui_rpc_auth.cfg. This has the key so you can remote connect to the Linux BOINC client via BOINC Manager.

Attach to the Seti project and give BOINC a minute to create the project directory. Then stop BOINC. Extract the Seti archive with the app_info.xml file to the project directory at BOINC/projects/setiathome.berkeley.edu. Start the BOINC client again.

So...Tell me what you think I should do. There was a thread about this same topic a while ago. In my case, I installed a fresh system on the machine and installed BOINC through the Ubuntu Software center. I got version 7.0.27 again, but, it actually worked fine this time. There was a time when the version from the center would immediately error out the tasks as they downloaded. Just like last *times*, I got the notice that there was a new version out and I should download it. I have done this before and never could get past the 'just double click it' part. When you do get past the double click, you end up with a new BOINC folder in your Home folder, and the old one in the default location.

To make it short, I finally got the 7.0.28 version in my home folder working and nuked the Client folder lurking in the default location harbouring version 7.0.27. Things appeared to be working fine, but, I just got this in the event window;
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_01
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/stat_icon
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_02
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_00

I'm not sure if this is a major problem or not. I suppose I could just whip out my trusty USB pendrive and start over...again, but, things really seem to be working well this time and I'd rather avoid that if possible.
ID: 1316154 · Report as offensive
Profile David Anderson (not *that* DA) Project Donor
Avatar

Send message
Joined: 5 Dec 09
Posts: 215
Credit: 74,008,558
RAC: 74
United States
Message 1316194 - Posted: 16 Dec 2012, 19:44:24 UTC

The Ubuntu 12.04 boinc was broken, as you said.
The 12.10 Ubuntu version is a boinc 7.0.27 which works.

When you get the boinc distro from Berkeley really working,
could you post, here, a description of all you did and
where the files are and who boinc-manager and boinc-client run as
(as distinct from where they are in the Ubuntu version running as user boinc)?

I've been thinking about doing what you are doing but
was a bit reluctant to take the time, and I do really like
having boinc running as another user and group (with therefore
no permission to read my personal files because of how I set
umask).
ID: 1316194 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1316210 - Posted: 16 Dec 2012, 20:48:27 UTC - in response to Message 1316194.  

The Ubuntu 12.04 boinc was broken, as you said.
The 12.10 Ubuntu version is a boinc 7.0.27 which works.

When you get the boinc distro from Berkeley really working,
could you post, here, a description of all you did and
where the files are and who boinc-manager and boinc-client run as
(as distinct from where they are in the Ubuntu version running as user boinc)?

I've been thinking about doing what you are doing but
was a bit reluctant to take the time, and I do really like
having boinc running as another user and group (with therefore
no permission to read my personal files because of how I set
umask).

I'm one of the last persons to give advice on getting Linux to work, and can't really remember all the details. I had the version from the repository working, downloaded the update from Berkeley, moved it to my Home folder, and started working on trying to get it to work. I used the Software Center to remove the version they installed, then used the center to install it again. When I tried running the Home version files in the terminal I would get messages that the files libwxbase, libwxgtk, and libxssl were missing. When I tracked down the file locations I found they were not missing. I almost gave up...again. The biggest thing is to set the files to 'Allow executing file as program' and set the other permissions. Since the New BOINC Folder is in my Home Folder, I'm listed as the Owner/User and don't need to use gksu nautilus to view the project folder. I have BOINC in the launchpad, but, I have to double click /home/tbar/BOINC/boincmgr to get it to launch. It might stop working on the next reboot for all I know. I guess I'll just keep using it and see what happens.
ID: 1316210 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1316301 - Posted: 17 Dec 2012, 2:45:29 UTC - in response to Message 1316154.  
Last modified: 17 Dec 2012, 3:03:44 UTC


Nice info, I'll be looking at this later when I try to get my NV8800 working in 64bit Ubuntu. I have a couple of related questions. Could you explain how to install the BOINC 7.0.28 download SETI suggests you download and install after installing 7.0.27? I've not been able to install the downloaded .28 version and have it work without also having .27 'installed'. Assume that you've installed .27 in it's default location and you want to install .28 in your Home Folder. That's about where I am at present. Also, would you know where to find the package for 'Linux 64 bit AK v8.0 SSSE3x'? That App appears to have fallen off the face of the Earth. I must have downloaded the 32bit version the last time I was contemplating running SETI on Ubuntu, I have a copy of that App but desire the 64bit version. I've been looking at these threads, Can't get GPU to process workunits on Linux machine and Ubuntu 12.04 and GPU NVIDIA while trying to set up my newest SETI machine.
Thanks.


This BOINC install should be fairly simple. You can grab any version of BOINC which will be an executable file. Run that file, and the BOINC files automatically extra to a directory called BOINC. Here is a breakdown on how to do this in a shell:

wget http://boinc.berkeley.edu/dl/boinc_7.0.28_x86_64-pc-linux-gnu.sh
chmod 755 boinc_7.0.28_x86_64-pc-linux-gnu.sh
./boinc_7.0.28_x86_64-pc-linux-gnu.sh
cd BOINC
./boinc --allow_remote_gui_rpc (Start the BOINC client)

In the BOINC directory, there is a file called gui_rpc_auth.cfg. This has the key so you can remote connect to the Linux BOINC client via BOINC Manager.

Attach to the Seti project and give BOINC a minute to create the project directory. Then stop BOINC. Extract the Seti archive with the app_info.xml file to the project directory at BOINC/projects/setiathome.berkeley.edu. Start the BOINC client again.

So...Tell me what you think I should do. There was a thread about this same topic a while ago. In my case, I installed a fresh system on the machine and installed BOINC through the Ubuntu Software center. I got version 7.0.27 again, but, it actually worked fine this time. There was a time when the version from the center would immediately error out the tasks as they downloaded. Just like last *times*, I got the notice that there was a new version out and I should download it. I have done this before and never could get past the 'just double click it' part. When you do get past the double click, you end up with a new BOINC folder in your Home folder, and the old one in the default location.

To make it short, I finally got the 7.0.28 version in my home folder working and nuked the Client folder lurking in the default location harbouring version 7.0.27. Things appeared to be working fine, but, I just got this in the event window;
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_01
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/stat_icon
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_02
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_00

I'm not sure if this is a major problem or not. I suppose I could just whip out my trusty USB pendrive and start over...again, but, things really seem to be working well this time and I'd rather avoid that if possible.

I'm still receiving those errors every hour or so and have developed a working theory. Who/Whatever is looking for those files is probably looking for them in the folder that I nuked in the default location. It was basically a 0.27 duplicate of the 0.28 BOINC folder now in my Home Folder. Those files are in my Home BOINC folder and working fine when I use the simple view of the Manager. That is also probably why when I lock the icon to the launchpad it doesn't launch the program after exiting. It is also looking for the nuked folder in the default location. Now all I have to do is fix that. Everything else appears to be working normally.

I was just trying to update 0.27. Here are the instructions I was following;
After downloading BOINC you must install it:

Save the file to disk.
Double-click on the file icon.

It wasn't that easy :-)
ID: 1316301 · Report as offensive
HankB

Send message
Joined: 1 Dec 04
Posts: 12
Credit: 6,115,914
RAC: 1
United States
Message 1316667 - Posted: 18 Dec 2012, 2:31:38 UTC - in response to Message 1315894.  

Many thanks for the detailed reply.
In order for BOINC to detect NVIDIA GPUs in Linux, two things are necessary. One is libcuda.so and the other is the kernel module nvidia.ko. If the distribution is 64-bit, this library could be in either /usr/lib64 or /usr/lib. If you are trying to run 32-bit BOINC in 64-bit Linux, then you will also need the 32-bit NVIDIA libraries including libcuda.so installed which is a question that the NVIDIA driver installer asks. The 32-bit libraries may be in /usr/lib or /usr/lib32 depending on distribution.

hbarta@olive ~ $ find  /usr/lib -name libcuda.so
/usr/lib/nvidia-current/libcuda.so
/usr/lib/nvidia-experimental-310/libcuda.so

hbarta@olive ~ $ ls -l /usr/lib/nvidia-current/libcuda.so /usr/lib/nvidia-experimental-310/libcuda.so
lrwxrwxrwx 1 root root 17 Aug  5 10:40 /usr/lib/nvidia-current/libcuda.so -> libcuda.so.295.40
lrwxrwxrwx 1 root root 17 Oct 16 17:19 /usr/lib/nvidia-experimental-310/libcuda.so -> libcuda.so.310.14

hbarta@olive ~ $ find /usr/lib -name libcuda.so.310.14
/usr/lib/nvidia-experimental-310/libcuda.so.310.14

hbarta@olive ~ $ which boinc
/usr/bin/boinc
hbarta@olive ~ $ file /usr/bin/boinc
/usr/bin/boinc: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked (uses shared libs), for GNU/Linux 2.6.24, BuildID[sha1]=0xe40d23c766ec3359f299ae346cc59f24ca21176f, stripped
hbarta@olive ~ $ file /usr/lib/nvidia-current/libcuda.so.295.40 
/usr/lib/nvidia-current/libcuda.so.295.40: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, stripped
hbarta@olive ~ $ file /usr/lib/nvidia-experimental-310/libcuda.so.310.14 
/usr/lib/nvidia-experimental-310/libcuda.so.310.14: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, stripped
hbarta@olive ~ $ 


It looks like the libraries are present, right architecture and in the path configured for ld.


To check if nvidia.ko is loaded:
/sbin/lsmod | grep nvidia

hbarta@olive ~ $ lsmod|grep nvidia
nvidia               9367655  64 


Present and loaded.


The Seti application also depends on two NVIDIA libraries including libcudart.so and libcufft.so. These should come with the application but are otherwise available with the NVIDIA CUDA toolkit. When in doubt, you can run this command on the application to see if all necessary libraries are in place:

ldd setiathome_x41g_x86_64-pc-linux-gnu_cuda32 (replace with the name of the application that you are running)
I'll pursue this once I get boinc to recognize the GPU.

I'm presently runing bionc 7.0.27 from the Mint repository (actually Ubuntu 12.04)
hbarta@olive ~ $ dpkg -l | grep boinc
ii  boinc                                  7.0.27+dfsg-5ubuntu0.12.04.1            metapackage for the BOINC client and the manager
ii  boinc-client                           7.0.27+dfsg-5ubuntu0.12.04.1            core client for the BOINC distributed computing infrastructure
ii  boinc-manager                          7.0.27+dfsg-5ubuntu0.12.04.1            GUI to control and monitor the BOINC core client
ii  boinc-nvidia-cuda                      7.0.27+dfsg-5ubuntu0.12.04.1            metapackage for CUDA-savvy BOINC client and manager


Suggestions on what to check next are appreciated.

thanks,
hank

ID: 1316667 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 1316717 - Posted: 18 Dec 2012, 6:27:02 UTC

Do you have the proprietary Nvidia GPU drivers installed ? These are separate to the CUDA libraries

CUDA will not work with the Nouveau drivers.

T.A.
ID: 1316717 · Report as offensive
Jeroen
Volunteer tester

Send message
Joined: 28 Mar 01
Posts: 13
Credit: 8,155,940
RAC: 0
United States
Message 1316934 - Posted: 19 Dec 2012, 1:20:47 UTC - in response to Message 1316154.  


Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_01
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/stat_icon
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_02
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_00

I'm not sure if this is a major problem or not. I suppose I could just whip out my trusty USB pendrive and start over...again, but, things really seem to be working well this time and I'd rather avoid that if possible.


Were the permissions changed on the projects folder? From reading the log, it looks like the BOINC client does not have permission to write to that directory. Do you know what user the BOINC client is running as?
ID: 1316934 · Report as offensive
Jeroen
Volunteer tester

Send message
Joined: 28 Mar 01
Posts: 13
Credit: 8,155,940
RAC: 0
United States
Message 1316939 - Posted: 19 Dec 2012, 1:33:19 UTC - in response to Message 1316667.  


hbarta@olive ~ $ ls -l /usr/lib/nvidia-current/libcuda.so /usr/lib/nvidia-experimental-310/libcuda.so
lrwxrwxrwx 1 root root 17 Aug 5 10:40 /usr/lib/nvidia-current/libcuda.so -> libcuda.so.295.40
lrwxrwxrwx 1 root root 17 Oct 16 17:19 /usr/lib/nvidia-experimental-310/libcuda.so -> libcuda.so.310.14

hbarta@olive ~ $ find /usr/lib -name libcuda.so.310.14
/usr/lib/nvidia-experimental-310/libcuda.so.310.14




Can you try creating a symbolic link for libcuda.so in /usr/lib?

ln -s /usr/lib/nvidia-experimental-310/libcuda.so.310.1 /usr/lib/libcuda.so

Also, check to see if nouveau is loaded.

lsmod |grep nouveau

If this module is loaded, you will either need to uninstall nouveau drivers or blacklist this kernel module.
ID: 1316939 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1316941 - Posted: 19 Dec 2012, 1:44:29 UTC - in response to Message 1316934.  
Last modified: 19 Dec 2012, 1:45:14 UTC


Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_01
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/stat_icon
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_02
Sun 16 Dec 2012 01:10:05 PM EST | SETI@home | [error] Can't create link file projects/setiathome.berkeley.edu/slideshow_setiathome_enhanced_00

I'm not sure if this is a major problem or not. I suppose I could just whip out my trusty USB pendrive and start over...again, but, things really seem to be working well this time and I'd rather avoid that if possible.


Were the permissions changed on the projects folder? From reading the log, it looks like the BOINC client does not have permission to write to that directory. Do you know what user the BOINC client is running as?

Yes, it appeared to be a permission problem with the slideshow files. About a day ago I went through the folder permissions again and set them to R&W, R, R. I also changed the permissions on those files themselves. They were set as owner-root. I set them to TBar R&W, R, R and the errors went away never to return. There are other files there with root as owner but they are not giving any problems. The only problem left is the launchpad not launching, but, I don't spend all day launching the program so it's not a biggie. It's probably another permissions issue. Does Ubuntu have a program such as Apple Disk Utility that Automagically corrects the permissions? It would be nice.

I need to get this board in a case so I can install the NV8800 to work MB tasks. With the NV8800 & AMD6850 in the same box the scheduler is constantly running the 8800 out of work and I'm tired of dealing with it. It was great when we had AstroPulse tasks. The NV8800 worked on MBs, the 6850 worked strictly APs. Now with both doing MB tasks, it's hopeless. We need the APs back.
ID: 1316941 · Report as offensive
HankB

Send message
Joined: 1 Dec 04
Posts: 12
Credit: 6,115,914
RAC: 1
United States
Message 1317307 - Posted: 19 Dec 2012, 20:20:03 UTC - in response to Message 1316939.  


Can you try creating a symbolic link for libcuda.so in /usr/lib?

ln -s /usr/lib/nvidia-experimental-310/libcuda.so.310.1 /usr/lib/libcuda.so

Also, check to see if nouveau is loaded.

lsmod |grep nouveau

If this module is loaded, you will either need to uninstall nouveau drivers or blacklist this kernel module.

The symlink solved the problem with the GPU not being recognized. The nouveau module was not loaded.

Wed 19 Dec 2012 11:27:28 AM CST |  | NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 5.0, compute capability 2.1, 134214655MB, 134214350MB available, 907 GFLOPS peak)
Wed 19 Dec 2012 11:27:28 AM CST |  | OpenCL: NVIDIA GPU 0: GeForce GTX 460 (driver version 310.14, device version OpenCL 1.1 CUDA, 1023MB, 134214350MB available)


thanks,
hank

ID: 1317307 · Report as offensive
Profile Ex: "Socialist"
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 3433
Credit: 2,616,158
RAC: 2
United States
Message 1319776 - Posted: 25 Dec 2012, 2:27:54 UTC - in response to Message 1317307.  


Can you try creating a symbolic link for libcuda.so in /usr/lib?

ln -s /usr/lib/nvidia-experimental-310/libcuda.so.310.1 /usr/lib/libcuda.so

Also, check to see if nouveau is loaded.

lsmod |grep nouveau

If this module is loaded, you will either need to uninstall nouveau drivers or blacklist this kernel module.

The symlink solved the problem with the GPU not being recognized. The nouveau module was not loaded.

Wed 19 Dec 2012 11:27:28 AM CST |  | NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 5.0, compute capability 2.1, 134214655MB, 134214350MB available, 907 GFLOPS peak)
Wed 19 Dec 2012 11:27:28 AM CST |  | OpenCL: NVIDIA GPU 0: GeForce GTX 460 (driver version 310.14, device version OpenCL 1.1 CUDA, 1023MB, 134214350MB available)


thanks,
hank

I've had to symlink other lib.so files from the nvidia folder for other issues in Linux Mint, mainly with the proprietary driver being used the system still wanted to use noveau's GLX module instead of nvidia's. Seems to be a common problem currently and not just for GLX apparently.
#resist
ID: 1319776 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Linux GPU setup


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.