CUDA, Linux Ubuntu 10.04 and strange mismatch version

Accelerated Computing CUDA CUDA Programming and Performance
21

Hi,

… a bug ?

http://forums.nvidia.com/index.php?showtopic=181448

Norge

22

Dear BenjaminG,

I have the same error. I think that it is because the mother-board does not recognize the NVIDIA card.

I changed to another mother-board, and I installed UBUNTU 10.10.
At now, the C2050 was recognized and driver 260.24 was installed in a good way.
After that, I installed CUDA 3.2 version, as well as, the corresponding SDK examples.
and … IT IS RUNNING !!!

Norge

23

Dear BenjaminG,

I have the same error. I think that it is because the mother-board does not recognize the NVIDIA card.

I changed to another mother-board, and I installed UBUNTU 10.10.
At now, the C2050 was recognized and driver 260.24 was installed in a good way.
After that, I installed CUDA 3.2 version, as well as, the corresponding SDK examples.
and … IT IS RUNNING !!!

Norge

24

I have the same error as the OP,

I am trying to use,

CUDA driver: devdriver_3.2_64_260.24

CUDA toolkit: 3.2.9_64

CUDA SDK: 3.2

Linux: Debian Lenny 64 bit

GCC: 4.3.2

I previously followed this guide

http://forums.nvidia.com/lofiversion/index.php?t171590.html

In particular the post by slaupin_bauks (Jun 30 2010, 06:38 PM)

It worked fine for 3.1 but for 3.2 I also had to add

NVCCFLAG := --compiler-options -fpermissive

to common.mk in order to get the SDK to compile.

Now I cannot get any SDK example or my own code to run unless I am root.

Interestingly, I also cannot run nvclock unless I am root.

Unfortunately, gthazmatt’s workaround does not work for me. I still cannot run anything as a normal user after I run deviceQuery as root.

Can you all stop quoting the entire OP please? It makes the thread unreadable and achieves nothing :S

Also, Norge the card is being recognised. Everything works with previous versions of CUDA.

EDIT

My humble apologies, I had a problem with my nis groups where non-local users weren’t being put in the local video group correctly. My CUDA 3.2 works fine now that I’ve fixed this. I hope you guys fix your problem as well.

25

I have the same error as the OP,

I am trying to use,

CUDA driver: devdriver_3.2_64_260.24

CUDA toolkit: 3.2.9_64

CUDA SDK: 3.2

Linux: Debian Lenny 64 bit

GCC: 4.3.2

I previously followed this guide

http://forums.nvidia.com/lofiversion/index.php?t171590.html

In particular the post by slaupin_bauks (Jun 30 2010, 06:38 PM)

It worked fine for 3.1 but for 3.2 I also had to add

NVCCFLAG := --compiler-options -fpermissive

to common.mk in order to get the SDK to compile.

Now I cannot get any SDK example or my own code to run unless I am root.

Interestingly, I also cannot run nvclock unless I am root.

Unfortunately, gthazmatt’s workaround does not work for me. I still cannot run anything as a normal user after I run deviceQuery as root.

Can you all stop quoting the entire OP please? It makes the thread unreadable and achieves nothing :S

Also, Norge the card is being recognised. Everything works with previous versions of CUDA.

EDIT

My humble apologies, I had a problem with my nis groups where non-local users weren’t being put in the local video group correctly. My CUDA 3.2 works fine now that I’ve fixed this. I hope you guys fix your problem as well.

26

I have a similar situation running on a headless Ubuntu 10.10 (64-bit) server/processor:

devdriver_3.2_linux_64_260.19.21.run
cudatoolkit_3.2.16_linux_64_ubuntu10.04.run
gpucomputingsdk_3.2.16_linux.run

Since I’m running headless, I have no GUI–no gdm

I’ve noticed that after installing the above packages, there are no nvidia devices listed in /dev

$ ls /dev | grep nvidia

$ cat /proc/devices | grep nvidia
195 nvidia

$ lspci | grep nVidia
01:00.0 VGA compatible controller: nVidia Corporation GF104 [GeForce GTX 460] (rev a1)
01:00.1 Audio device: nVidia Corporation GF104 High Definition Audio Controller (rev a1)

According to the quick-start, I need to set up an init script to use mknod. Apparently there is another way, though.

For some reason, compiling the SDK examples and running:
$ sudo ./C/bin/linux/release/deviceQueryDrv
{blah}
actually makes the correct device nodes. Afterward, I can run other examples without sudo.

I figured this might be a related `issue’ – I’ll put the mknod calls in an init script once I figure out the proper way to do that in ubuntu. (I hear you are supposed to us upstart instead of /etc/init.d/ …)

27

I have a similar situation running on a headless Ubuntu 10.10 (64-bit) server/processor:

devdriver_3.2_linux_64_260.19.21.run
cudatoolkit_3.2.16_linux_64_ubuntu10.04.run
gpucomputingsdk_3.2.16_linux.run

Since I’m running headless, I have no GUI–no gdm

I’ve noticed that after installing the above packages, there are no nvidia devices listed in /dev

$ ls /dev | grep nvidia

$ cat /proc/devices | grep nvidia
195 nvidia

$ lspci | grep nVidia
01:00.0 VGA compatible controller: nVidia Corporation GF104 [GeForce GTX 460] (rev a1)
01:00.1 Audio device: nVidia Corporation GF104 High Definition Audio Controller (rev a1)

According to the quick-start, I need to set up an init script to use mknod. Apparently there is another way, though.

For some reason, compiling the SDK examples and running:
$ sudo ./C/bin/linux/release/deviceQueryDrv
{blah}
actually makes the correct device nodes. Afterward, I can run other examples without sudo.

I figured this might be a related `issue’ – I’ll put the mknod calls in an init script once I figure out the proper way to do that in ubuntu. (I hear you are supposed to us upstart instead of /etc/init.d/ …)