Installation

Requirements

Hardware

NVIDIA GPU is required to run EPicker. The following GPUs have been tested.

• NVIDIA RTX 2080Ti
• NVIDIA Tesla P100
• NVIDIA Tesla V100
• NVIDIA Tesla A100
• NVIDIA Titan X

System

EPicker requires a linux envirement, and has been tested on the following systems.

• CentOS 7
• Ubuntu 18.04
• Ubuntu 16.04

EPicker is developed using python, supposed to be able to run on MS Windows or MAC, but is never tested.

GCC and CUDA

Pytorch is used in the deep learning framework. Because a specified version of Pytorch is not compatible with all CUDA versions, we provide multiple installation packages for different environments. The EPicker package for CUDA9.0 requires g++ 4.8.5 or newer. For the ohter packages require higher CUDA version, g++ 5.0 or newer are needed.

Network

If the machine to install EPicker doesn't have internet access, the offline packages are needed. All third party dependencies are included.

If your machine has the internet access, you can use the online packages which will take some tiemes to download all the required depndencies.

Installation

The installation of EPicker should be carried out on a machine with the same GPU to be used, for example, on a computing node with GPU, because Pytorch needs to detect the type of GPU and compile with according parameteres. This problem is inconvenient and expected to be solved in the future.

The installation is quite simple, directly run the package script to install as following:

$ cd {Path to the installation folder}  
$ ./EPicker-x.x.x-cuxx-xxline.sh

The installation paths of CUDA and g++ will be asked during the installation. The version of the provided CUDA should match the CUDA version labeled in the package filename.