Howtoinstall Cuda,cudnn,tf

How to benchmark GPUs using Tensorflow

Environment

• OS : ubuntu 16.04.5 (4.15.0-29-generic)
• gpu : nvidia gtx 1070 mxm
  • driver : 410.72
  • cuda : 9.0.176
  • cuDNN : cuda-9.0-linux-x64-v7.4.1.5
• python : 3.5.2
  • pip : 18.1
  • tensorflow : tf-nightly-gpu 1.13.0.dev20181122

 

Disable nouveau

#!/bin/bash
CONF=/etc/modprobe.d/blacklist-nouveau.conf

echo "blacklist nouveau" >> $CONF
echo "options nouveau modeset=0" >> $CONF
update-initramfs -u
reboot

From askubuntu.com

 

Install cuda-9.0

cuda include the nvidia driver, so you do not need to install driver separately.

root@localhost# wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
root@localhost# dpkg -i cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
root@localhost# apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
root@localhost# apt-get update
root@localhost# apt-get install cuda-9-0

Add envs to ~/.bashrc

export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64

 

Install cuDNN-7.4

Download cudnn-9.0-linux-x64-v7.4.1.5.tgz at developer.nvidia.com
And perform following commands

root@localhost# tar xfz cudnn-9.0-linux-x64-v7.4.1.5.tgz
root@localhost# cp -P cuda/include/* /usr/local/cuda/include/
root@localhost# cp -P cuda/lib64/* /usr/local/cuda/lib64/

 

Install Tensorflow

Do not install tensorflow-gpu, it is not compatible with tf_cnn_benchmarks.py

root@localhost# python3 -m pip install tf-nightly-gpu

 

Tensorflow-Benchmarks

Download scripts

root@localhost# wget https://github.com/tensorflow/benchmarks/archive/master.zip

Run scripts

#!/bin/bash

PY=/usr/bin/python3
SCRIPTS=/root/benchmarks-master/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py

${PY} ${SCRIPTS} --num_gpus=1 --batch_size=32 --model=resnet50  --variable_update=parameter_server --local_parameter_device=cpu

refer to github