deviceQuery on Tesla fails

OS : Red Hat Enterprise Linux Server 5.6 (64 bit)

GPU : nVidia Tesla M2050

SDK : GpuComputingSDK_3.2.16_linux

Toolkit : CudaToolkit_3.2.16_linux_64_rhel5.5

I have no root access and won`t get one. No physical access to machine either (e.g. for a BIOS or so). I could request a restart, hopefully.

This is what I get when I execute deviceQuery:

[font=“Courier New”]

bash-3.2$ ./deviceQuery

./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount FAILED CUDA Driver and Runtime version may be mismatched.

FAILED

Press to Quit…


[/font]

Now, since I don`t have root access and default destination at home directory has disk quotas, I had to change it, which gives me SDK compilation issues:

[font=“Courier New”]/usr/bin/ld: skipping incompatible /opt/local/glut/libglut.so when searching for -lglut

/usr/bin/ld: cannot find -lglut

collect2: ld returned 1 exit status

make[1]: *** […/…/bin/linux/release/nbody] Error 1

make[1]: Leaving directory `/visuals_local/NVIDIA_GPU_Computing_SDK/C/src/nbody’

make: *** [src/nbody/Makefile.ph_build] Error 2[/font]

I looked into release notes and it states that I should perform a symbolic link. However, I dont have write permissions in the directory where the original glut (libglut.so.3) resides. These are the things Ive tried:

  • copy & rename the glut .so files into new directory and make symbolic link into new dir

  • set CPATH (there is even glut.h inside the SDK) and LD_LIBRARY_PATH environment variables to point to locations where the glut is

  • make -L

It simply can`t find the glut at other locations.

Is it possible to skip SDK examples compilation process and at least try creating my own examples ? I`d guess not, since I get the above error from deviceQuery.

Any ideas ?

Thanks

We managed to get the IT support execute the magic command , so now the SDK is compiled successfully.

However, we still get this error message:

[font=“Courier New”]./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount FAILED CUDA Driver and Runtime version may be mismatched.

FAILED

Press to Quit…


[/font]

Any ideas ?

Thanks.

What driver version is installed?

im having troubles running anything with user-level access too; but i do have root access yet the users cant run code…

This is the scenario:

  1. The user “guy” logs in, and tries to run either deviceQuery or any other code from the SDK samples and it fails.
  2. The root “god” logs in, and tries to run either deviceQuery or any other code from the SDK samples and it runs succesfully.
  3. “guy” can now run any code.

So, the only way a user can run any code is if root runs any code before.

helpo?

I get the same error.

Additionally when I try running ./clock I get:

xxxxxx@xxxxxxxx:~/NVIDIA_GPU_Computing_SDK/C/bin/linux/release$ ./clock
clock.cu(73) : cudaSafeCall() Runtime API error : no CUDA-capable device is detected.

System specs:

OS: Ubuntu Server 10.04
GPU: Tesla S2050

Update: sudo ./deviceQuery works! After I copied the libs needed to /usr/lib. After the sudo any user can execute deviceQuery.