AI-based applications require a lot of computing power, and as these applications get faster, better, and smarter, chipmakers are racing to develop the next generation of powerful AI chips. One firm that’s making waves in the AI market and competing with big names like NVIDIA and Qualcomm is Graphcore, a UK-based firm that has recently closed its latest round of funding for a total of $222 million.
What are Intelligence Processing Units?
Graphcore, which is led by the former CEO of XMOS, Nigel Toon, specializes in intelligence processing units (IPU). This is a type of flexible parallel processor for artificial and machine intelligence workloads. According to Graphcore, IPUs are designed to do for AI what the graphics processing unit (GPU) did for computing.
Indeed, GPUs have been used for some time in artificial neural networks (ANNs), the human brain-inspired computing systems that are able to recognize relationships between large data sets. However, some modern ANNs now include millions of parameters in models, which may necessitate a new computing method.
Graphcore’s ColossusTM Mk2 GC200 IPU. Image used courtesy of Graphcore
“We need to move towards compute that’s a lot more sparse to support much larger models,” says Toon. Instead, he suggests that engineers need to design entirely new machines to do the job—current GPUs and GPUs won’t cut it.
A Stride Toward Autonomy
Graphcore’s IPUs are currently used in robotics and driverless cars. The chips are designed to make AI more efficient by using multiple cores to improve the natural language processing (NLP) power of systems. In 2018, Microsoft Azure turned to Graphcore’s IPU. Then this past summer, the startup released its second generation of the technology.
The second-generation IPU integrates 1,472 separate IPU cores and is capable of carrying out 8,832 separate parallel computing threads. Each IPU processor receives a performance boost from a set of floating-point technologies unique to Graphcore chips, called AI-Float.
Competing with Industry Mainstays
The power of Graphcore’s IPU technology gives the company the potential to compete with industry leaders like NVIDIA and Qualcomm. With the industry growing more competitive, the E-series funding comes at a good time and will help the company grow and innovate further.
The company has also built-in dedicated AI networking into each IPU-M2000 called IPU-Fabric. Image used courtesy of Graphcore
This year, NVIDIA overtook Intel to become the most valuable chipmaker in the United States. Then, in September, NVIDIA announced plans to acquire Arm from SoftBank, which was met with criticism from Toon as “bad for competition” and the market.
Following the completion of the funding round, he said in a statement, “Having the backing of such respected institutional investors says something very powerful about how the markets now view Graphcore”.
From Funding to Hardware Development
The E-series funding brings the total raised since Graphcore’s founding in 2016 to $710 million, valuing the company at $2.77 billion, up from $1.5 billion two years ago.
Given that tech firms like Apple, Amazon, Microsoft, and Google are all gearing up to make their own AI chips—indeed, Graphcore sells its chips through partners including Microsoft—this funding round could be critical for Graphcore’s future growth plans, particularly given that it’s trying to position itself as a key player in increasing AI adoption within emerging technologies like 5G and the cloud.