Calzone Deep Learning Setup

gpu_setup.md

Calzone Deep Learning Setup

• Ubuntu 18.10
• NVIDIA Titan X

$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 18.10
Release:        18.10
Codename:       cosmic

$ lspci -vnn | grep VGA
01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP104 [GeForce GTX 1080] [10de:1b80] (rev a1) (prog-if 00 [VGA controller])
02:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP102 [TITAN X] [10de:1b00] (rev a1) (prog-if 00 [VGA controller])

Install Steps

1. Install nvidia driver (latest)

sudo add-apt-repository ppa:graphics-drivers/ppa
ubuntu-drivers autoinstall

2. Install CUDA (ver 10.0.130)

$ wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-repo-ubuntu1804_10.0.130-1_amd64.deb
$ sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
$ sudo dpkg -i cuda-repo-ubuntu1804_10.0.130-1_amd64.deb
$ sudo apt update
$ sudo apt install cuda-10-0

3. Install cuDNN (7.5.0 for CUDA 10.0) https://developer.nvidia.com/rdp/cudnn-download

$ dpkg -i libcudnn7_7.5.0.56-1+cuda10.0_amd64.deb
$ dpkg -i libcudnn7-dev_7.5.0.56-1+cuda10.0_amd64.deb
$ dpkg -i libcudnn7-doc_7.5.0.56-1+cuda10.0_amd64.deb

4. Install OpenBLAS

$ aptitude install libopenblas-dev

5. Install Bazel https://github.com/bazelbuild/bazel/releases

• Select 0.21.0 for tensorflow r1.13 or latest for master

$ https://github.com/bazelbuild/bazel/releases/download/0.21.0/bazel_0.21.0-linux-x86_64.deb
$ dpkg -i bazel_0.21.0-linux-x86_64.deb

6. Install Tensorflow

$ git clone https://github.com/tensorflow/tensorflow.git
$ cd tensorflow
$ git checkout r1.13
$ python3 -m virtualenv venv -p python3
$ source venv/bin/activate
$ pip install six mock numpy keras_applications keras_preprocessing

• Configuration
  • When prompted for CUDA support: Y
  • When prompted for gcc: /usr/bin/gcc-7

$ ./configure

• Build

$ bazel build --config=opt --config=cuda --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" //tensorflow/tools/pip_package:build_pip_package

• Wait several hours
• If using r1.13:

$ ./bazel-bin/tensorflow/tools/pip_package/build_pip_package ~

• If using master:

$ ./bazel-bin/tensorflow/tools/pip_package/build_pip_package --nightly_flag ~

• Cleanup

$ rm -rvf ~/.cache/bazel
$ git clean -df
$ rm bazel-bin bazel-genfiles bazel-out bazel-tensorflow bazel-testlogs