If you reboot teslas or fermis, be sure to run the initialization script:

 sudo nvidia-smi -a 

Most recent cuda & sdk installation (applicable to tesla1~tesla4 & fermi2)

cuda4.1:

/opt/nvidia/cuda41: cuda toolkit

/opt/nvidia/cuda41/NVIDIA_GPU_Computing_SDK: cuda sdk

legacy ( stable ) cuda & sdk installation (applicable to tesla1~tesla4 & fermi1 ~ fermi4 )

cuda4.0:

/opt/nvidia/cuda4: cuda toolkit

/opt/nvidia/cuda4/NVIDIA_GPU_Computing_SDK: cuda sdk

-Note, if using the 'cutil' library, the flag is now -lcutil_x86_64 and not -lcutil, as it was for 2.3

cuda2.3:

/opt/nvidia/cuda: cuda toolkit

/opt/nvidia/cuda-2.3_sdk: cuda sdk

The working MPI is in /opt/MPI/mpich2-1.2.1-install/

All the includes/libs/executables are in there, and to use it with fortran, make sure you have the fortran compiler

/opt/intel/Compiler/11.1/059/bin/intel64/bin/ifort