Graphcore

Graphcore
source
Name	Graphcore
Type	Private
Industry	Semiconductor
Founded	2016
Founders	Nigel Toon, Simon Knowles
Headquarters	Bristol, England
Key people	Nigel Toon, Simon Knowles
Products	Intelligence Processing Unit (IPU), IPU-Machine
Website	company website

Graphcore

Graphcore is a British semiconductor company developing accelerators and systems for artificial intelligence and machine learning workloads. The company designs the Intelligence Processing Unit (IPU) architecture and associated hardware and software to compete with offerings from NVIDIA, Intel Corporation, AMD, Google (company), and Amazon (company). Graphcore's work has positioned it within the broader landscape of Deep learning, High-performance computing, and specialist processor design firms such as Cerebras Systems and SambaNova Systems.

History

Graphcore was founded in 2016 by Nigel Toon and Simon Knowles after their prior ventures and collaborations in networking and semiconductor companies. Early investors included figures and institutions connected to SoftBank Group, Sequoia Capital, and Atomico, and funding rounds placed Graphcore among leading European startups alongside DeepMind, ARM Holdings, and Ocado Group. The company began product development amid competition from Intel Corporation acquisitions such as Nervana Systems, the emergence of Google TPU, and advances at NVIDIA with its CUDA-driven GPU ecosystems. Over time Graphcore announced partnerships with research institutions including University of Cambridge, University of Oxford, and corporate collaborations with Microsoft, Dell Technologies, and Hewlett Packard Enterprise.

Technology

Graphcore's core innovation is the IPU, a processor architecture optimized for fine-grained, low-latency matrix and graph computations characteristic of modern neural networks. The IPU departs from x86 and ARM-centric paradigms by emphasizing thousands of independent compute cores, on-chip memory, and high-bandwidth interconnects inspired by techniques seen in Chiplet design and custom interposers used by firms like TSMC and Intel Foundry Services. Graphcore's architecture targets models used in Transformer (machine learning model), Convolutional neural network, and reinforcement learning research exemplified by work at OpenAI, DeepMind, and academic labs at Stanford University and Massachusetts Institute of Technology. The IPU hardware integrates with network fabrics and rack-scale deployments similar to designs from Cray Research-era supercomputing and modern clusters from Hewlett Packard Enterprise and Dell Technologies.

Products

Graphcore has released a series of products including IPU chips and systems such as the IPU-Machine and IPU-M2000 series aimed at datacenter and enterprise customers. These systems are positioned against competitors' accelerators like the NVIDIA A100, Google TPU v4, and custom ASICs by Apple Inc. and Amazon Web Services. Graphcore's offerings include scale-up and scale-out configurations for training and inference workloads used by companies in sectors represented by Facebook (Meta Platforms), Google (company), Microsoft, and research groups at University College London and ETH Zurich. Accessories and integrations include PCIe cards, mezzanine modules, and cloud-hosted instances provided by partners such as Microsoft Azure and select service providers.

Software and Ecosystem

Graphcore provides a software stack centered on Poplar, a graph-programming framework and SDK designed to map machine learning workloads to IPU hardware. Poplar interoperates with common frameworks and formats like PyTorch, TensorFlow, and ONNX to enable model migration from environments used by teams at OpenAI, Google DeepMind, and academic labs such as University of California, Berkeley. The company fosters an ecosystem of tooling, profilers, and libraries comparable to CUDA Toolkit, cuDNN, and vendor software from Intel such as oneAPI and MKL. Graphcore also emphasizes compiler technology, scheduling, and memory management techniques similar to advances from compiler projects at Google and research groups at Carnegie Mellon University.

Business and Funding

Graphcore's financing rounds attracted venture capital from notable firms including Sequoia Capital, Atomico, and strategic investors tied to SoftBank Group and sovereign investment funds also involved with European technology ventures. The company navigated valuation events amid market shifts affecting peers such as Cerebras Systems and SambaNova Systems, and sought revenue growth through partnerships with hyperscalers and enterprises like Microsoft and Dell Technologies. Graphcore has also pursued manufacturing relationships with foundries and supply chain partners including TSMC and logistics providers used by global semiconductor companies like Foxconn.

Reception and Impact

Graphcore has been recognized in technology press and analyst reports alongside NVIDIA, Intel Corporation, and startup rivals for advancing alternative processor architectures for artificial intelligence, with commentators referencing comparisons to historic moves by ARM Holdings and the rise of domain-specific architectures noted in publications discussing Moore's law-era transitions. Research groups at institutions including Stanford University, Massachusetts Institute of Technology, and University of Cambridge have examined IPU performance on benchmarks drawn from work by OpenAI and DeepMind. Critics and proponents alike discuss trade-offs versus established GPU ecosystems from NVIDIA and accelerator offerings from Google (company) and Amazon Web Services in terms of software maturity, developer adoption, and total cost of ownership.

Category:Semiconductor companies