CUDA Driver Version is insufficient for CUDA Runtime CUDA Error Runtime

I am trying to run some CUDA Codes on Tesla (S1070) Workstation; (I am not the admin)

I have successfully compiled them (using a Make File & Manually) using CUDA 3.1 and I get an error
cudaSafeCallNoSync() Runtime API error : CUDA driver version is insufficient for CUDA runtime version.

I even downgraded to CUDA 3.0 and at runtime I get the same error.

The source Code is on
http://people.maths.ox.ac.uk/~gilesm/cuda/
Try to run the 3rd Example (Scan Reduction)
I am able to run the 1st & 4th examples.
(2nd example needs some external libraries)

Please help me in this for this is very important for me.
Thanks in Advance.
Cheers
Abhishek :(

Forgot to mention that the OS is Scientific Linux 5.4 on a National GPU Computing Facility with 4 Tesla (S1070) cards on each node with GCC 4.1.2