Advertisement
Advertisement
CNBC
Get more with myNEWS
A personalised news feed of stories that matter to you
Learn more
Pilot models of the Uber self-driving car are displayed at the Uber Advanced Technologies Centre in the US. Angelo Merendino/AFP

Samsung-backed start-up making chips for AI uses raises US$30 million to take on Intel

British firm says its processors will be up to 100 times better than those on the market

CNBC

A computer chip start-up designing hardware for artificial intelligence (AI) applications has raised US$30 million to help it take on giants like Intel and Nvidia .

UK-based Graphcore is aiming to ship its chips next year to be used in applications from driverless cars to cloud computing, processes that will require machine-learning technology. This involves large amounts of data being used to teach AI systems.

Graphcore said its processors will be up to 100 times better than current solutions in the market when it comes to how quickly the AI can learn, as well as the amount of energy consumed.

Currently, Graphic Processing Units (GPUs) are used to run machine-learning applications. But Nigel Toon, the chief executive of Graphcore said these are not suitable for the future.

“GPUs have been built to run programmes that completely describe the algorithm. Machine learning is different. You are trying to teach the system using data and that requires a different style of compute,” Toon told CNBC.

Graphcore is creating what it calls an Intelligent Processing Unit (IPU) system to market in 2017.

The funding round was led by Robert Bosch Venture Capital and included Samsung, as well as venture capital firms Amadeus Capital Partners, C4 Ventures, Draper Esprit, Foundation Capital and Pitango Venture Capital.

Graphcore has been working on its technology for the past two years and Toon claims that the technology will be able to do help reduce data centre costs and improve efficiencies. Toon gave the example of a social media company that typically has the most users are at certain times of the day. But when user levels are low, the servers which would have Graphcore’s IPU, could use the downtime to train AI systems, then deploy that update the following day.

The market is potentially huge, with spending on hardware for deep learning projects set to grow from US$436 million in 2015 to US$41.5 billion by 2024, according to marketing intelligence firm Tractica.

But at the moment, that market is dominated by incumbents Nvidia and Intel. Both companies earlier this year unveiled next generation processors specifically designed for running artificial intelligence applications.

Initally, the start-up is aiming to sell to firms that are looking to train AI systems but the company said it is already talking to its strategic investors about the next wave of applications.

“We have Bosch as a strategic investor, we have Samsung as an investor and Bosch is interested in autonomous vehicles and the next generation of transportation, while Samsung is interested in missing word edge of network devices,” Toon said.

“We are working with partners on some of these other applications.”

Post