RL @illya13

Install Kubeflow on a single-node Kubernetes v1.18.2 cluster

Intro

This installation is based on:

• kfctl v1.0.2 - control plane for deploying and managing Kubeflow
• kfctl_k8s_istio config - vanilla deployment of Kubeflow with all its core components without any external dependencies
• single-node Kubernetes v1.18.2 cluster that was created in previous tutorial

Other goal is to have GPU support. So please check other tutorials:

• Installing TensorFlow 2 with GPU and Docker support on Ubuntu 20.04 LTS
• Installing PyTorch 1.5 with GPU and Docker support on Ubuntu 20.04 LTS

Installing Kubernetes with GPU support on Ubuntu 20.04 LTS

Intro

There are multiple ways to install Kubernetes on Ubuntu 20.04. The easiest but not the best way is to use MicroK8s for single node setup. At this point I would not recommend using MicroK8s if you need GPU support:

• I personally failed to make it working with GPU support
  • standard K8s GPU example doesn’t work
  • Revert to containerd 1.2.5 until we get gpu to work on 1.3.0
• it is trying to do too many things on multiple platforms
• snap-based installs are still questionable.

So we’ll go with Kubeadm install in this tutorial.

Installation Prerequisites

Before continuing Kubernetes installation process, please make sure you already have installed:

• NVIDIA GPU drivers
• docker: > 19.03
• NVIDIA Container Toolkit

Installing PyTorch 1.5 with GPU and Docker support on Ubuntu 20.04 LTS

Intro

According to https://pytorch.org/get-started/locally/ and https://github.com/NVIDIA/nvidia-docker the following has to be installed:

• python: >= 3.6
• python package managers: Anaconda or pip
• docker: > 19.03

NVIDIA software:

• NVIDIA GPU drivers
• CUDA Toolkit: 10.2
• NVIDIA Container Toolkit

Please also note:

• CUDA 10.2 requires GCC <= 8

Installing TensorFlow 2 with GPU and Docker support on Ubuntu 20.04 LTS

Intro

Please check Installing TensorFlow 2 with GPU support on Ubuntu 20.04 LTS for the intro.

According to https://www.tensorflow.org/install/docker and https://github.com/NVIDIA/nvidia-docker the following has to be installed:

• docker: > 19.03
• NVIDIA GPU drivers
• NVIDIA Container Toolkit

Docker in Ubuntu 20.04 LTS

There are number of docker packages available

• docker-ce package from docker.com
• docker.io package provided by Canonical
• docker package provided by Red Hat

Docker versions > 19.03 are supported by NVIDIA Container Toolkit.

Installing TensorFlow 2 with GPU support on Ubuntu 20.04 LTS

Intro

According to https://www.tensorflow.org/install and https://www.tensorflow.org/install/gpu the following has to be installed:

• python: 3.5 – 3.7
• pip: > 19.0

NVIDIA software:

• NVIDIA GPU drivers: > 418.x
• CUDA Toolkit: 10.1
• cuDNN SDK: >= 7.6

Please also note:

• CUDA 10.1 requires GCC <= 8

At time of writing the following versions were available:

• TensorFlow: 2.1.0
• NVIDIA GPU drivers: 440.82
• CUDA Toolkit: 10.1.243 (with GPU drivers 418.87)
• cuDNN SDK: v7.6.5