October 28–31, 2019
Please log in

Targeting high-performance ML accelerators using XLA

Victoria Rege (Graphcore), David Norman (Graphcore)
4:10pm4:50pm Wednesday, October 30, 2019
Location: Grand Ballroom H
Average rating: **...
(2.00, 1 rating)

Who is this presentation for?

  • Software developers




Victoria Rege and David Norman outline the best practices that have emerged as part of Graphcore’s integration process with Poplar. They detail the company’s insights and optimizations with the broader TensorFlow developer community so it can run its models built with TensorFlow more efficiently.

You’ll walk away knowing how to optimize new ML accelerators, in particular, the intelligence processing unit, and how to use the Poplar software stack and TensorFlow using the XLA backend.

Prerequisite knowledge

  • General knowledge of TensorFlow
  • Experience with GPUs

What you'll learn

  • Understand how the TensorFlow XLA backend works and how to use it to efficiently target advanced ML accelerators
Photo of Victoria Rege

Victoria Rege


Victoria Rege is the head of strategic partnerships at Graphcore, where she works with key customers and leads research and universities AI engagements. She has over a decade of experience in the semiconductor space. Previously, she held several leadership positions at NVIDIA from global alliances, product marketing, and campaigns to the founding of the GPU Technology Conference; and she has worked in the hedge fund space as executive director for the Hedge Fund Business Operations Association. Victoria is a frequent contributor to ACM SIGGRAPH and is AR, MR & VR Chair for the SIGGRAPH 2019 Conference. She’s also an active member of the Consumer Technology Association’s AI Working Group.

Photo of David Norman

David Norman


Dave Norman is the director of machine learning frameworks at Graphcore, where he heads the frameworks team and is the creator of the intelligence procession unit and Poplar Software. He’s been in software engineering for over 25 years, specializing in real-time, high-performance and embedded systems. Previously, he was at Hewlett Packard, writing control software for experimental wireless and broadband modems; has worked for various companies on drivers for novel hardware, 3G/4G base stations, and the tools chain for an FPGA-like architecture; and worked abroad in New Zealand developing real-time weather graphics.

  • O'Reilly
  • TensorFlow
  • Google Cloud
  • IBM
  • Databricks
  • Tensor Networks
  • VMware
  • Amazon Web Services
  • One Convergence
  • Quantiphi
  • Lambda Labs
  • Tech Mahindra
  • cnvrg.io
  • Determined AI
  • Inferencery
  • Manceps, Inc.
  • PerceptiLabs
  • Valohai

Contact us


For conference registration information and customer service


For more information on community discounts and trade opportunities with O’Reilly conferences


For information on exhibiting or sponsoring a conference


For media/analyst press inquires