Document Updates

The main updates to the documents are summarised below. In addition there will have been many incremental improvements and corrections.

November 2022

New/updated documents

• Updated and extended user guide for the IPU Operator integration with Kubernetes to coincide with the first public release (Using IPUs in Kubernetes)

September 2022

SDK 3.0 Release

For full details of all the changes in this release, see the SDK 3.0 Release Notes.

• New user guide for Model Runtime (Model Runtime: User Guide).

• The user guide and API reference for GCL has been moved to a separate document (GCL User Guide and API Reference).

Tutorials and examples

There are two new tutorials:

• A guide to help set up Visual Studio code with the Poplar SDK and IPUs.

• A tutorial which shows how to programmatically access PopVision trace instrumentation (PVTI) files.

July 2022

New/updated documents

• Added an alphabetical list of all documents.

• Updated Getting Started with Graphcloud.

SDK 2.6 Release

For full details of all the changes in this release, see the SDK 2.6 Release Notes.

• Targeting the IPU from TensorFlow 2 has been revised to reflect the move to TensorFlow 2.6.

• New user guides for TensorFlow Serving (IPU TensorFlow Serving 1 User Guide, IPU TensorFlow Serving 2 User Guide).

Tutorials and examples

• There is a new tutorial on training a Hugging Face model using a local dataset. The PyTorch example trains a Vision Transformer (ViT) model.

• There is a new series of tutorials introducing PopXL, an experimental Python module with which you can create models directly. This provides greater flexibility than is possible using the standard PopART API.

  • Tutorial 1: Basic Concepts.

  • Tutorial 2: Custom Optimiser.

  • Tutorial 3: Data Parallelism.

  • Tutorial 4: Pipelining.

  • Tutorial 5: Remote Variables and Replicated Tensor Sharding (RTS).

  • Tutorial 6: Phased Execution.

• There is a new PopVision tutorial on Lightweight profiling.

• A section on “Generating and profiling instant events” has been added to the PopVision tutorial Instrumenting applications.

• There is a new feature example demonstrating how to use recomputation checkpoints in TensorFlow 2.

• We removed the numbering from these tutorials as they can be followed in any order.

  • PyTorch:

    • Introduction to PopTorch.

    • Efficient data loading with PopTorch.

    • BERT Fine-tuning on IPU.

    • Half and mixed precision in PopTorch.

    • Observing tensors in PopTorch.

    • PopTorch Parallel Execution Using Pipelining.

    The Introduction to PopTorch tutorial has also been extended to include a section on importing your own image into the model.

  • PopVision:

    • Accessing profiling information.

    • Instrumenting applications.

June 2022

• Major new section on Scaling an application over multiple replicas in the technical note Memory and Performance Optimisation on the IPU.

• Poplar SDK release notes and TensorFlow installation documents updated with information about protobuf compatibility issue.

May 2022

New documents

• Added a set of Quick Starts to get started on Graphcloud.

• More reference architectures added: Graphcore IPU Based Systems with Weka Data Platform and Graphcore Pod with DDN storage describe the systems developed with Weka and DDN.

SDK 2.5 release

For full details of all the changes in this release, see the SDK 2.5 Release Notes.

• Targeting the IPU from TensorFlow 2 has been revised to reflect the move to TensorFlow 2.5.

Tutorials and examples

• There is a new tutorial demonstrating use of TensorBoard in TensorFlow 2 on the IPU (Using TensorBoard with TensorFlow 2 on the IPU).

• There is now a tutorial demonstrating parallel execution through pipelining using PyTorch (Parallel execution using pipelining).

• A new example shows how to use concurrent pipelining in TensorFlow 1 (Concurrent pipelining).

March 2022

• Addition of the documents for the Bow range of products.

• Extended IPU Programmer’s Guide with much new material.

Tutorials and examples

• The following tutorials are now available in Markdown and as Jupyter notebooks:

  • TensorFlow 1 Pipelining tutorial.

  • TensorFlow 2 Keras tutorial.

  • PyTorch tut2 (Efficient Data Loading).

• The following simple applications have been turned into tutorials, and are available in Markdown and as Jupyter notebooks:

  • TensorFlow 1 MNIST simple tutorial.

  • TensorFlow 2 MNIST simple tutorial.

  • PyTorch MNIST simple tutorial.

January 2022

• Scaling AI with Graphcore and Pure Storage describes an example reference architecture developed with Pure Storage.

December 2021

New documents

• The Memory and Performance Optimisation on the IPU guide describes general techniques for developing high-performance machine learning models running on the IPU.

• Graphcore OpenStack Reference Design for IPU-POD Systems illustrates a reference configuration of a Graphcore IPU‑POD₆₄ rack deployed with OpenStack.

SDK 2.4 release

For full details of all the changes in this release, see the SDK 2.4 Release Notes.

• New gcipuinfo library for monitoring and gathering information about the IPUs available in a system, and the applications using them.

Tutorials and examples

• There are two new examples demonstrating use of the Poplar distributed configuration library (PopDist) for:

  • distributed training of a TensorFlow 2 model using the PopDist API.

  • distributed training of a PyTorch model using the PopDist API.

• The following changes have been made to the Poplar tutorials:

  • Tutorial 5 has been moved to simple_applications/poplar/mnist.

  • Tutorials 6 & 7 have been re-numbered and can now be found in tut5_matrix_vector and tut6_matrix_vector_opt, respectively.

• The PyTorch tutorial on Fine-tuning BERT on the IPU is now available as a Jupyter notebook and a Python script.

• The following tutorials are now available in Markdown and as Jupyter notebooks:

  • PyTorch tut1 (Basics).

  • PyTorch tut3 (Half and mixed precision in PopTorch).

• The PopART pytorch_api MNIST application has been deleted because it is not compatible with the current version of PopART.

October 2021

SDK 2.3 release

For full details of all the changes in this release, see the SDK 2.3 Release Notes.

Tutorials and examples

• New TensorFlow 1 Pipelining Tutorial.

• The TensorFlow 2 Graph recompilation tutorial was removed, as similar content is available in the technical note on optimising for the IPU in TensorFlow. This now covers both TensorFlow 1 and TensorFlow 2.

• The Distributed MNIST with Horovod Training Demo was removed because it uses a feature (HostAllReduce) which is no longer supported. For distributed training, use PopRun and PopDist instead. The example is compatible with SDK 2.1 and SDK 2.2.

August 2021

SDK 2.2 release

• Targeting the IPU from TensorFlow 2 has been revised to reflect the move to TensorFlow 2.4

• The PopVision Analysis Library (libpva) User Guide and PopVision Trace Instrumentation Library documents have been split out from the Poplar & PopLibs User Guide

• The Vertex/Assembler Programming Guide has been merged into the Poplar and PopLibs User Guide

For full details of all the changes in this release, see the SDK 2.2 Release Notes.

New tutorials and examples

• Using Infeed and Outfeed Queues in TensorFlow 2

• Observing tensors in PyTorch

• Using a custom op in a PyTorch model

July 2021

A new white paper on Graphcore’s AI-Float technology with hardware support for mixed-precision and stochastic rounding.

• Mixed-Precision Arithmetic for AI: A Hardware Perspective

June 2021

SDK 2.1 release

For full details of all the changes in this release, see the SDK 2.1 Release Notes.

Other changes

• The Graphcore GitHub repositories have been split into a section for tutorials and examples to help you get started, and another with sample applications for the IPU.

• New tutorial on using the PopVision analysis library (libpva)

• New example of using PopDist for distributed training

• New technical note published on using the availableMemoryProportion option to optimise memory use: Optimising Temporary Memory Usage for Convolutions and Matmuls on the IPU

• Updated TensorFlow 1 tutorial on half precision training:

  • an improved section on loss scaling

  • a new section on how to avoid numerical issues (overflow, underflow, unstable operations)

  • and an additional example using the IPUEstimator

• Updated PyTorch tutorial on half precision training:

  • a new section on numerical stability (loss scaling, stochastic rounding)

  • a section on debugging floating-point exceptions

  • a section on PopTorch tracing

• Updated Poplar profiling tutorial (tutorial 4):

  • code for the IPU Model and for running on hardware (previously only the IPU Model could be used)

  • an improved section on out-of-memory errors

  • an improved section on navigating a profile generated by PopVision Graph Analyser

May 2021

• Published a new technical note on the implementation and performance of BERT-Large on Graphcore IPU-POD systems: Pre-Training and Fine-Tuning BERT for the IPU

• The V-IPU Administrator Guide has been updated with a new chapter on monitoring application metrics using Prometheus and InfluxDB.

April 2021

• A new technical note has been published with an overview of how to add custom operations in PyTorch, TensorFlow and PopART: Creating Custom Operations for the IPU.

• The technical note Model Parallelism on the IPU with TensorFlow: Sharding and Pipelining was revised to be consistent with SDK 2.0, and with the addition of content on pipelining Keras models in TensorFlow 2.

• The “Writing custom operations” chapter in the TensorFlow 1 and TensorFlow 2 user guides has been completely rewritten,

• New tutorials have been added on using half and mixed-precision floating point in PopTorch and TensorFlow 1, and efficient data loading with PopTorch

March 2021

Hardware documents

• The IPU-POD Direct Attach Build and Test Guide has been updated.

• Published the IPU-POD16 Direct Attach Datasheet.

Updates since SDK 2.0

Since the SDK release, the following documents have been updated:

• The Graphcore Command Line Tools document has been updated to 1.0.51. This adds a description of the information returned by the gc-info --tile-overview option and documents the new gc-exchangewritetest command.

• The Targeting the IPU from TensorFlow 1 has been updated with an improved explanation of the role of XLA.

SDK 2.0 release

For full details of all the changes in this release, see the SDK 2.0 Release Notes.

The main documentation changes are:

• Additions to:

  • Targeting the IPU from TensorFlow 1

  • Targeting the IPU from TensorFlow 2

  • PyTorch for the IPU: User Guide

• New: PopDist and PopRun: User Guide documents the tool for running distributed programs and the associated PopDist library.

• New and updated functions in the Poplar and PopLibs API Reference:

  • New libraries:

    • poplar/DebugContext.hpp

    • poplar/TensorRearranger.hpp

    • poplin/Cholesky.hpp

    • poplin/TriangularSolve.hpp

    • popnn/CTCLoss.hpp

    • popnn/LogSoftmax.hpp

    • popnn/Rnn.hpp

    • popops/CollectiveTypes.hpp

    • popops/SequenceSlice.hpp

    • popops/SortOrder.hpp

    • popops/TopK.hpp

    • popsparse/MatMulParams.hpp

    • poputil/OptionParsing.hpp

    • poputil/TensorMetaData.hpp

  • Changes:

    • Renamed poplar/CycleEstimateFunc.hpp to poplar/PerfEstimateFunc.hpp

    • Renamed popops/Collectives.hpp to popops/TensorCollectives.hpp

• The PopVision trace instrumentation (PVTI) and PopVision analysis (PVA) libraries are now included in the in Poplar docs.

  • PVTI User Guide

  • PVA C++ API

  • PVA Python API

  • PVTI C++ API

Other changes

• There is now a contents page listing the examples and tutorials available on GitHub.

• The IPU-M2000 DA system has been renamed to Getting Started with IPU-POD4 DA and IPU-POD16 DA and updated.

November 2020

This includes the release of Poplar SDK 1.4, the IPU-M2000, the IPU-POD reference design and the supporting software tools.

• Poplar SDK 1.4

  • Updates for new features in the SDK 1.4 release (see the SDK release notes for details of changes and new features in the SDK)

  • PopART: New chapter on writing custom ops

  • PopTorch: extended PyTorch for the IPU: User Guide & new GitHub tutorial & how-to video

  • Keras: Introduction to using Keras in TensorFlow 2 on the IPU

  • TensorFlow: Added documentation and GitHub tutorial on Graph Recompilation in TensorFlow 2 (note: this tutorial was removed with SDK 2.3 so is not guaranteed to be compatible with this or later releases)

  • PopVision User Guide for the System Analyser and the trace instrumentation library (libpvti)

  • Additional video tutorials: Using Data Feeds & Training Loops, Evaluating Batch Sizes for IPUs and Fundamentals of BSP

• New and updated documents for IPU-POD64 reference design

  • Getting Started with IPU-POD Systems

  • IPU-POD64 Data Centre Checklist

  • IPU-POD64 Reference Design: Build and Test Guide

  • IPU-POD64 Reference Design: Installation and Integration Guide

  • IPU-POD64 Reference Design Datasheet

• New documents for IPU-M2000 direct attach

  • Getting Started with IPU-POD4 DA and IPU-POD16 DA

  • IPU-POD Direct Attach Build and Test Guide

  • IPU-M2000 IPU-Machine Datasheet

• Documents for hardware support tools:

  • V-IPU Administrator Guide

  • V-IPU User Guide

  • BMC User Guide