Graphcore

Graphcore was founded in 2016 by Simon Knowles and Nigel Toon.{{cite news |url=https://www.theguardian.com/business/2020/dec/29/uk-graphcore-valued-28bn-nvidia-artificial-intelligence |title=UK chipmaker Graphcore valued at $2.8bn after it raises $222m |date=2020-12-29 |work=The Guardian |last=Jolly |first=Jasper}}

In the autumn of 2016, Graphcore secured a first funding round led by Robert Bosch Venture Capital. Other backers included Samsung, Amadeus Capital Partners, C4 Ventures, Draper Esprit, Foundation Capital, and Pitango.{{cite news| title = AI chipmaker Graphcore raises $30 million to take on Intel| author = Arjun Kharpal| url = https://www.cnbc.com/2016/10/31/ai-chipmaker-graphcore-raises-30-million-to-take-on-intel-nvidia.html| newspaper = CNBC| date = 2016-10-31| access-date = 2017-07-31}}{{cite news| title = UK chip start-up Graphcore raises £30m for take on AI giants| author = Madhumita Murgia| url = https://www.ft.com/content/053f9dea-9e98-11e6-891e-abe238dee8e2| newspaper = Financial Times| date = 2016-10-31| access-date = 2017-08-02}}

In July 2017, Graphcore secured a round B funding led by Atomico,{{cite news| title = U.K. Chip Designer Graphcore Gets $30 Million to Fund Expansion| author = Jeremy Kahn and Ian King| url = https://www.bloomberg.com/news/articles/2017-07-20/u-k-chip-designer-graphcore-gets-30-million-to-fund-expansion| newspaper = Bloomberg| date = 2017-07-20| access-date = 2017-07-31}} which was followed a few months later by $50 million in funding from Sequoia Capital.{{cite news|last1=Lynley|first1=Matthew|title=Graphcore raises $50M amid a flurry of AI chip activity |date = 2017-11-12| url=https://techcrunch.com/2017/11/12/graphcore-raises-50m-amid-a-flurry-of-ai-chip-activity/|access-date=2017-12-07 |work=TechCrunch|language=en}}

In December 2018, Graphcore closed its series D with $200 million raised at a $1.7 billion valuation, making the company a unicorn. Investors included Microsoft, Samsung and Dell Technologies.{{Cite web|url=https://techcrunch.com/2018/12/18/ai-chip-startup-graphcore-closes-200m-series-d-adds-bmw-and-microsoft-as-strategic-investors/|title=AI chip startup Graphcore closes $200M Series D, adds BMW and Microsoft as strategic investors|website=TechCrunch|date=18 December 2018 |language=en-US|access-date=2018-12-19}}

On 13 November 2019, Graphcore announced that their Graphcore C2 IPUs were available for preview on Microsoft Azure.{{Cite web|url=https://www.graphcore.ai/posts/microsoft-and-graphcore-collaborate-to-accelerate-artificial-intelligence|title=Microsoft and Graphcore collaborate to accelerate Artificial Intelligence|last=Toon|first=Nigel|website=www.graphcore.ai|language=en-gb|access-date=2019-11-16}}

Meta Platforms acquired the AI networking technology team from Graphcore in early 2023.{{cite news |last=Paul |first=Katie |date=5 May 2023 |title=Meta Platforms scoops up AI networking chip team from Graphcore |url=https://www.reuters.com/technology/meta-platforms-scoops-up-ai-networking-chip-team-graphcore-2023-05-05/ |publisher=Reuters}}

In July 2024, Softbank Group agreed to acquire Graphcore for around $500 million. The deal is under review by the UK's Business Department's investment security unit.{{Cite web |last=Nicol-Schwarz |first=Kai |date=9 July 2024 |title=Graphcore employees have share value wiped as sale to SoftBank agreed |url=https://sifted.eu/articles/graphcore-conditional-sale-agreed-news |work=Sifted}}{{Cite web |last1=Titcomb |first1=James |last2=Field |first2=Matthew |date=1 July 2024 |title=Japanese deal for AI champion Graphcore faces national security review |url=https://www.telegraph.co.uk/business/2024/07/01/softbank-ai-deal-graphcore-national-security-review/ |work=The Daily Telegraph}}

In 2016, Graphcore announced the world's first graph tool chain designed for machine intelligence called Poplar Software Stack.{{Cite web|url=https://www.graphcore.ai/posts/what-does-machine-learning-look-like|title=Inside an AI 'brain' - What does machine learning look like?|last=Fyles|first=Matt|website=www.graphcore.ai|language=en-gb|access-date=2019-11-16}}{{Cite web|url=https://www.graphcore.ai/posts/introducing_poplar_our_ipu_processor_software_at_neurips|title=Introducing Poplar® - our IPU-Processor software at NeurIPS|last=Doherty|first=Sally|website=www.graphcore.ai|language=en-gb|access-date=2019-11-16}}{{Cite web|url=https://www.graphcore.ai/posts/graph-computing-for-machine-intelligence-with-poplar|title=Graph computing for machine intelligence with Poplar™|last=Fyles|first=Matt|website=www.graphcore.ai|language=en-gb|access-date=2019-11-16}}

In July 2017, Graphcore announced its first chip, called the Colossus GC2, a "16 nm massively parallel, mixed-precision floating point processor", that became available in 2018.{{cite web|url=https://www.hpcwire.com/2017/07/20/graphcore-readies-launch-16nm-colossus-ipu-chip/|title=Graphcore Readies Launch of 16nm Colossus-IPU Chip |last1=Trader |first1=Tiffany |date=2017-07-20 |website=hpcwire.com |publisher=HPC Wire |access-date=2017-12-11}}{{cite web|url=https://silvertonconsulting.com/blog/2018/11/19/new-graphcore-gc2-chips-with-2pflop-performance-in-a-dell-server/|title=New GraphCore GC2 chips with 2PFlop performance in a Dell Server |last1=Lucchesi |first1=Ray |date=2018-11-19 |website=silvertonconsulting.com |publisher=Silverton Consulting |access-date=2018-12-16}} Packaged with two chips on a single PCI Express card, called the Graphcore C2 IPU (an Intelligence Processing Unit), it is stated to perform the same role as a GPU in conjunction with standard machine learning frameworks such as TensorFlow. The device relies on scratchpad memory for its performance rather than traditional cache hierarchies.{{cite web|url=https://www.graphcore.ai/hubfs/assets/pdf/Citadel%20Securities%20Technical%20Report%20-%20Dissecting%20the%20Graphcore%20IPU%20Architecture%20via%20Microbenchmarking%20Dec%202019.pdf|title=Dissecting the Graphcore IPU Architecture via Microbenchmarking|author=Citadel High Performance Computing R&D Team|date=2019}}

In July 2020, Graphcore presented its second generation processor called GC200, built with TSMC's 7nm FinFET manufacturing process. GC200 is a 59 billion transistor, 823 square-millimeter integrated circuit with 1,472 computational cores and 900 Mbyte of local memory.{{Cite web|url=https://www.graphcore.ai/posts/introducing-second-generation-ipu-systems-for-ai-at-scale/|title=Graphcore Introducing 2nd Generation IPU Systems For AI At Scale|language=en-gb|access-date=2020-08-09}} In 2022, Graphcore and TSMC presented the Bow IPU, a 3D package of a GC200 die bonded face to face to a power-delivery die that allows for higher clock rate at lower core voltage.Timothy Prickett Morgan: [https://www.nextplatform.com/2022/03/03/graphcore-goes-3d-with-ai-chips-architects-10-exaflops-ultra-intelligent-machine/ GraphCore Goes Full 3D With AI Chips.] The Next Platform, March 3, 2022. Graphcore aims at a Good machine, named after I.J. Good, enabling AI models with more parameters than the human brain has synapses.

Both the older and newer chips can use 6 threads per tile{{what?|date=July 2024}} (for a total of 7,296 and 8,832 threads, respectively) "MIMD (Multiple Instruction, Multiple Data) parallelism and has distributed, local memory as its only form of memory on the device" (except for registers).{{cn|date=July 2024}} The older GC2 chip has 256 KiB per tile while the newer GC200 chip has about 630 KiB per tile that are arranged into islands (4 tiles per island),{{Cite arXiv|title=Dissecting theGraphcore IPUArchitecturevia Microbenchmarking|eprint=1912.03413|last1=Jia|first1=Zhe|last2=Tillman|first2=Blake|last3=Maggioni|first3=Marco|author4=Daniele Paolo Scarpazza|year=2019|class=cs.DC }} that are arranged into columns, and latency is best within tile.{{what?|date=July 2024}}{{cn|date=July 2024}} The IPU uses IEEE FP16, with stochastic rounding, and also single-precision FP32, at lower performance.{{Cite web|title=THE GRAPHCORE SECOND GENERATION IPU|url=https://www.graphcore.ai/hubfs/MK2-%20The%20Graphcore%202nd%20Generation%20IPU%20Final%20v7.14.2020.pdf?hsLang=en}} Code and data executed locally must fit in a tile, but with message-passing, all on-chip or off-chip memory can be used, and software for AI makes it transparently possible,{{what?|date=July 2024}} e.g. has PyTorch support.{{cn|date=July 2024}}

{{portal|Companies}}

• AI accelerator
• Massively parallel processor array

{{Reflist}}

• [https://www.graphcore.ai/ Graphcore]

{{differentiable computing}}

{{coord|51|27|19.0|N|2|35|33.3|W|type:landmark|display=title}}

Category:64-bit computers

Category:British brands

Category:Semiconductor companies of the United Kingdom

Category:British companies established in 2016

Category:Companies based in Bristol

Category:Fabless semiconductor companies

Category:2024 mergers and acquisitions