Help with enabling Tensorflow GPUs with Amazon EC2 - P2 Instance

I need help with the installation process for Nvidia drivers and cuda toolkit that support an Amazon EC2 - P2 Instance. I need this to run tensorflow libraries that are enabled by the Amazon provided GPU. I am experiencing “Failed to initialize NVML: Driver/library version mismatch” when I perform nvidia-smi.

Amazon P2 instances use K80 GPUs, and it claims to have kernel version 384.81 which would correspond to the Cuda 9.0 driver. I have tried installation guides for TensorFlow for Ubuntu 16.04 Cudas 10.1 and 9.0 here, but both are incompatible with the current kernel.

Can anyone offer a recommendation on cuda drivers and toolkit? Is there anything special about a setup in an Amazon cloud environment?

Please note, I am a novice in solving cuda, tensorflow, and linux issues.

Thanks,

LT