16-01-2021 | | By Robin Mitchell
Recently, a Bristol-based AI company raised $222 million in a Series E funding round to expand and grow its products. Who is Graphcore, what products do they produce, and what does this funding round demonstrate?
Graphcore is a Bristol-based company that specialises in AI processing systems. Conceived in 2012 and founded in 2016, the company started from a concept of hardware specialised in running AI systems. Through various funding, rounds have grown into a company with a net worth over $1.7 billion. The products developed by Graphcore have shown to be critical for deep learning and other AI tasks, and now serve multiple customers, including Microsoft Azure, Dell, and Cirrascale. The sudden growth of Graphcore has seen international interest in the company, and Graphcore now has offices in London, Cambridge, Oslo, Palo Alto, Beijing, and Taiwan.
What is an IPU?
While Graphcore creates a range of different products, they are centred around the Graphcore IPU which stands for intelligence processing unit. Graphcore has established a goal whereby all AI processing should be moved to its own dedicated core that would reside in a processor, and this core is referred to as an IPU.
Graphcore, being a semiconductor developer at heart, has already developed and deployed their IPU solution in the Colossus MK2 GC200 IPU form. This device integrates 1472 independent IPU cores with 8832 independent program threads all executing in parallel. Furthermore, the IPU integrates 900MB in-processor memory and has over 47 TB/s of memory bandwidth per IPU. Graphcore integrates these processors into PCIe cards that allow any computing system to take advantage of the IPU without needing a custom computer or motherboard.
Graphcore products are mainly aimed at mainframes, servers, and supercomputers, which are often required to execute large AI systems. For example, the Graphcore IPU-Machine is an AI building block that can act as a stand-alone system or be used on a network with other machines and has 1 PetaFlop of AI capability. Another example product produced by Graphcore is the IPU-Server which utilises standard OEM systems but integrates Graphcore dual-IPU PCIe cards.
The growing importance of AI and its increased use in multiple industries means that the demand for better AI systems has exponentially grown. Such is evident with Graphcore as they recently entered their Series E funding, which raised over $222 million.
The funding round led by Ontario Teachers Pension Plan Board has had funds added to the group via Fidelity International and Schroders. Combining this funding round with previous rounds, Graphcore has now raised more than $710 million and expects to have over $440 million in cash to continue its growth and develop future products.
“Having the backing of such respected institutional investors says something compelling about how the markets now view Graphcore. The confidence that they have in us comes from our competence in building our products and our business. We have created a technology that dramatically outperforms legacy processors such as GPUs, a powerful set of software tools tailored to the needs of AI developers, and a global sales operation that is bringing our products to market. With advances in artificial intelligence moving apace, we look forward to more innovation, enabled and powered by Graphcore technology.” - Nigel Toon, Co-Founder and CEO of Graphcore
While Graphcore products are aimed at the extreme end of AI, the use of dedicated hardware to execute AI has been in development by many other companies for the past decade. AI is a processor and memory heavy task due to the constraints of generic CPUs. However, specialised hardware that can execute many simple operations simultaneously (such as that found on GPUs) can run AI systems far more quickly and efficiently.
Since AI is becoming more mainstream, the need for everyday computing systems to run AI tasks is increasing, and the move away from cloud-based services favouring privacy could accelerate this need. As processor technology improves, designers may find themselves with excessive CPU power. Thus, chip designers could begin to cut back on CPUs' capabilities in favour of dedicated hardware circuits for complex tasks.