2. Software installation
In order to run models on a Pod system, you will need to download and install the following software packages from the Graphcore Downloads portal:
Poplar SDK which includes development tools and also command line tools for managing the IPU hardware (Section 2.1, Installing the Poplar SDK). You will also need to install framework-specific Python wheel files (included in the SDK) if you wish to run applications in PyTorch or TensorFlow on the IPU.
User-level V-IPU software (Section 2.2, Installing the V-IPU command-line tools).
2.1. Installing the Poplar SDK
Download the Poplar SDK tarball for your OS from the Graphcore Downloads portal.
You can check which OS you are running by using the command:
$ lsb_release -a
The tarball is named as follows:
poplar_sdk-[os]-[poplar_ver]-[build].tar.gz
where [os] is the host OS, [poplar_ver] is the SDK software version number and [build] identifies a specific build.
An example of the Poplar SDK tarball for Ubuntu 20.04 and version 3.2 of the SDK is:
poplar_sdk-ubuntu_20_04-3.2.0-7cd8ade3cd.tar.gz
Install the SDK by unpacking the tarball as follows:
$ tar -xvzf poplar_sdk-[os]-[ver].tar.gz
You can install the SDK to any directory that you have write access to.
For details of the components included in the Poplar SDK, refer to the Contents of the SDK section in the Poplar SDK Overview document.
2.2. Installing the V-IPU command-line tools
Note
A Pod DA system does not need the V-IPU software. If you are using a Pod DA, you can skip this section.
Note
You can omit this step if vipu is already installed on the system.
You can check if it is installed by running:
$ vipu --version
Download the user V-IPU tarball from the Graphcore Downloads portal.
Note
The V-IPU software can be installed on any computer that can communicate with the V-IPU controller that is running on a host server in the Pod.
The tarball is named as follows:
vipu_-[ver].[arch].tar.gz
where [ver] is the V-IPU software version number and [arch] is the architecture.
An example of the tarball for V-IPU version 1.18.0 is:
vipu-1.18.0.x86_64.tar.gz
Extract the contents of the tarball with:
$ tar xzvf vipu_-[ver].[arch].tar.gz
The tarball is extracted to a directory named:
$ vipu_-[ver]
Next add the directory to the $PATH environment variable:
$ export PATH=$PWD/vipu-[ver]:$PATH
For example, to add the directory for V-IPU version 1.16.1 to the $PATH environment variable:
$ export PATH=$PWD/vipu-1.18.0:$PATH
Confirm that that the vipu executable is found and that it reports the expected version, by running:
$ vipu --version
This should output the version of the software that you downloaded.
If the correct version is not reported, confirm that the V-IPU directory has been added to the to the $PATH environment variable with:
$ echo $PATH
2.3. Installations for PyTorch and TensorFlow
To run applications using PyTorch, PyTorch Geometric or TensorFlow, you will also need to install the Python wheel files (included in the Poplar SDK).
Details are given in the following Quick Start guides: