WHY THIS MATTERS IN BRIEF
The future will be powered by AI and Machine Learning, and it could also be powered by a new type of chip, an “Intelligence Processing Unit,” which is the next evolutionary step after the GPU.
What comes after the CPU and the GPU? The IPU. That’s if a whole bunch of tech giants and experts are to be believed anyway after a major Silicon Valley investor identified Graphcore, a small company based out of Bristol in the UK, as the world’s leading chip designer for Machine Learning (ML) and Artificial Intelligence (AI), backing them with $50m in a third round of funding.
If you’re still sceptical about the claims though well Graphcore might be different, and ARM’s Founder, Hermann Hauser, agrees, saying that the company “might well become the processor of choice for machine learning,” because unlike their predecessors these new chips, and their new revolutionary architecture allow AI’s to learn in a fundamentally new way, by learning from experience, just like us humans.
In short the new chip is akin to a human brain, a point that is best illustrated by using computational graphs to map what’s going on inside it. As you can see from the pictures below when you visualise what’s going on in there it looks remarkably like the human brain and all its connections…
As a result I’m sticking the company and their IPU tech, which I’ll come to in a moment, on my watch list, and now yours too. The investment from Sequoia Capital sees leading venture capitalists join the Graphcore board and technical advisory panel, and Sequoia has previously backed several companies that you might also know, including Apple, Cisco, Google, and NVIDIA.
If you have the time it’s worth a watch (13:00)
“Machine intelligence will cause an explosion of new applications and services that will transform every industry. We believe Graphcore’s product architecture, team and early market interest make it the best positioned new entrant in this market,” said Matt Miller, the partner at Sequoia who will join the Graphcore board of directors.
The deal brings the total investment in the company to over $110m, one of the largest in the field. The funding will be used to scale up production of what Graphcore call their Intelligence Processing Unit (IPU), which is an evolutionary leap forwards from the CPU and GPU’s we know today, chips and accelerator cards, building a community of developers around its Poplar ML software platform, driving Graphcore’s extended product roadmap, and investing in its Palo Alto based US team to help support customers.
The company is planning to ship its first IPU devices to early access customers at the start of 2018, slightly later than its original plan of the end of 2017. Early benchmarking has shown 10x to 100x speed up in running AI algorithms and the ability to scale across many separate accelerator cards.
“Efficient AI processing power is rapidly becoming the most sought-after resource in the technological world,” said Nigel Toon, CEO at Graphcore. “We believe our IPU technology will become the worldwide standard for machine intelligence compute. The performance of Graphcore’s processor, compared to other accelerators, is going to be transformative, whether you are a medical researcher, roboticist, online marketplace, social network or building autonomous vehicles. Sequoia has a deep understanding of what it takes to scale and build a successful business and we are looking forward to a long partnership with them.”
Last month, Graphcore shared preliminary benchmarks demonstrating that its IPU can improve performance of machine intelligence training and inference workloads by ten times to a hundred times compared with current hardware. One test showed that a developer could use eight IPU cards to run a training model in the same amount of time as 128 GPU cards. And that right there, dear readers, is the game changer.
Graphcore has already attracted investments from many of the biggest names in machine intelligence including Demis Hassabis, co-founder of revolutionary AI company DeepMind, Zoubin Ghahramani of University of Cambridge and Chief Scientist at Uber and Pieter Abbeel, Greg Brockman, Scott Grey and Ilya Sutskever, from OpenAI, which itself was backed by Elon Musk. And if you’re asking who any of those people are then, frankly, you’ve probably been asleep for the past decade.
The Series C round also has full support from Graphcore’s existing investors Amadeus Capital Partners, which is run by Hermann, the founder of ARM, Atomico, Robert Bosch Venture Capital, C4 Ventures, Dell Technologies Capital, Draper Esprit, Foundation Capital, Pitango Venture Capital and Samsung Catalyst Fund.
Miller will join Graphcore’s board of directors and Bill Coughran, another partner at Sequoia, will join Graphcore’s technical advisory board.