Instructions for compiling and running GPU code on cycle servers:
/usr/local/cuda-7.0/bin/nvcc -ccbin /usr/local/gcc-4.9.3/bin/ gpucode.cu

Instructions for compiling and running GPU code on the ITTC Slurm cluster (from login1):
1. Add CUDA to your environment:  module load CUDA/8.0.44
2. Compile gpucode.cu as follows:  nvcc -o gpucode gpucode.cu
[Here gpucode is the executable, and gpucode.cu is the CUDA source code.]
3. Write submission script following my example.  Call it gpuscript.sh
4. Submit job to cluster:  sbatch gpuscript.sh

#!/bin/bash
#SBATCH -p gpu
#SBATCH -c 1
#SBATCH --mem=4GB
#SBATCH --reservation=eecs739
#SBATCH --gres="gpu:p100:1"
#SBATCH -t 00:20:00
#SBATCH -J gpu_test
#SBATCH -o slurm-%j.out
echo "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"

# Setup environment
module load CUDA/8.0.44

# Run gpucode 
./gpucode