The latest release of Graphcore’s Poplar SDK, version 2.2, is now available.
Continuing our commitment to improved ease-of-use, expanded feature set, and performance enhancements, Poplar 2.2 includes a number of important updates.
For TensorFlow users, we have added preview support for TensorFlow 2.4, as well as refinements to our Keras implementation.
PyTorch users benefit from Torch 1.9 support, additional operators and support for PyTorch Lightning.
Poplar 2.2 also sees the release of new versions of Graphcore’s hugely popular Graph Analyser and System Analyser – collectively known as PopVision. These easy-to-use, visual tools give users a powerful insight into activity across their IPU system, allowing the optimisation of resources such as memory, code execution steps, CPU-IPU communication, and much more.
For full details of all the changes in this release, see the SDK 2.2.0 Release Notes.
Additionally, we are able to share updated information on performance for popular machine learning models, enabled by SDK 2.2.
As with each interaction of Poplar, we see significant performance gains. The following chart details enhancements made since SDK 1.4 in December 2020.
Poplar SDK documentation, revised in line with V2.2 updates are:
New tutorials and examples
A number of new tutorial and feature examples, relating to Poplar 2.2 are available on Graphcore’s GitHub.