SAN FRANCISCO — Graphcore, a developer of processors for machine learning and artificial intelligence, secured $50 million in additional funding, bringing the total raised by the UK-based startup to about $110 million over the past 18 months.
Graphcore's series C funding, provided by venture firm Sequoia Capital, will be used to scale up production of the startup's first chip, which it calls an Intelligence Processing Unit (IPU). Graphcore plans to make the IPU available to early access customers at the beginning of next year.
In addition to scaling up production, the new funding will be used to help build a community of developers around Graphcore's Poplar software platform, driving the company's extended product roadmap and investing in its Palo Alto, Calif.-based U.S. team to help support customers, Graphcore (Bristol, U.K.) said.
"Efficient AI processing power is rapidly becoming the most sought-after resource in the technological world," said Nigel Toon, Graphcore's CEO, in a press statement. "We believe our IPU technology will become the worldwide standard for machine intelligence compute."
Graphcore, which is featured in the most recent edition of EE Times Silicon 60, is perhaps the furthest along of a crop of startups that have been formed to create new processor architectures for deep neural networks (DNNs). In addition to Graphcore, other well-funded startups include Wave Computing, Cerebras and Groq.
These startups and others are in the early days of battle with the likes of more established companies such as Google, which has offered its Tensor Processing Unit (TPU) custom ASIC for machine learning since last year, Nvidia GPUs and Intel, which acquired Nervana and plans to sample its Neural Network Processor ASSP next year.
Nigel Toon
Toon maintains that the performance of Graphcore’s processor "is going to be transformative" compared to other accelerators. The company last month shared preliminary benchmarks that it says demonstrate that the IPU can improve performance of machine intelligence training and inference workloads by 10-100 times compared with current hardware.
Previous investors in Graphcore include both Samsung Catalyst Fund, the venture capital arm of Samsung, and Robert Bosch Venture Capital.
Matt Miller, a partner at Sequoia, will join Graphcore's board of directors as the result of the funding, the company said. Bill Coughran, another partner at Sequoia, will join Graphcore's technical advisory board, the company said.
— Dylan McGrath is the editor-in-chief of EE Times.
Related content:
- Big Data Reshapes Silicon
- Architectures Battle for Deep Learning
- Startup Schools Machine Learning
- AI Chip Startup Shares Insights