Meta has hired a team specializing in AI network chips and is developing an AI supercomputing system

According to news on May 6, Meta, the parent company of Facebook, recently hired a professional team of artificial intelligence network chips. The team had long been developing artificial intelligence network technology for British chip unicorn company Graphcore until the end of last year.

Meta spokesman Jon Carvill has confirmed the hiring. The media confirmed that 10 people on the team updated their profiles on LinkedIn, showing that they worked at Graphcore until December 2022 or January 2023, and then joined Meta in February or March this year.

Cavell said: “We have recently recruited a number of highly specialized engineers in Oslo to join Meta’s infrastructure team. They have deep expertise in designing and developing supercomputing systems to support Meta. Artificial Intelligence and Machine Learning in the Data Center The move brings additional support to Meta in improving how its data centers handle artificial intelligence efforts, as the company grapples with the fundamentals of artificial intelligence across teams across the company. Demand for the facility surged, and these teams were looking to develop new features. In particular, Meta-owned Facebook and Instagram increasingly rely on artificial intelligence technology to target ads, recommend posts and remove banned content.

Most importantly, Meta now joins competitors such as Microsoft and Google in hoping to launch innovative artificial intelligence products. The products promise to be able to generate text, art and other content that resembles real-person creations, an area investors see as the next big growth area for tech companies.

The job description on LinkedIn shows that this team of 10 employees has been engaged in research on artificial intelligence-specific network technology at Graphcore and is committed to developing computer chips and systems optimized for artificial intelligence work. Cavill declined to say what kind of work they would be doing at Meta.

According to a spokesperson for Graphcore, the company announced a restructuring in October last year and closed its Oslo office. At the time, the company was competing with U.S. companies such as Nvidia and Advanced Micro Devices, which were dominating the artificial intelligence chip market.

According to two sources, Meta has set up a dedicated team internally and designed several chips aimed at accelerating and maximizing the efficiency of its artificial intelligence work. These include a networking chip that performs an air traffic control-like function for servers.

Modern artificial intelligence systems, such as the ChatGPT chatbot and Dall-E image generation tool, require efficient network support. These systems are too large to fit on a single computing chip and must be spread across many chips connected in series.

To keep data flowing smoothly across these computing clusters, a new network chip has emerged. Companies such as Nvidia, AMD and Intel all produce such network chips.

In addition to network chips, Meta is also designing a complex computing chip for training artificial intelligence models and performing inference. Inference refers to the trained model making judgments and generating responses, but the chip is expected to take It won't be ready until around 2025.

Graphcore is one of the UK’s most highly valued tech startups. It was once regarded as a potential challenger to Nvidia by investors such as Microsoft and venture capital firm Sequoia, which may shake the latter's dominant position in the artificial intelligence chip system market. However, Graphcore suffered setbacks in 2020. At that time, Microsoft canceled an early deal to buy Graphcore chips for its Azure cloud computing platform. Instead, Microsoft used Nvidia's GPUs to build the large-scale infrastructure that supports ChatGPT developer OpenAI. Microsoft is also a major supporter of OpenAI.

Sequoia Capital has since written down its investment in Graphcore to zero, according to a person familiar with the matter. Although it still retains a seat on Graphcore's board of directors.

A spokesperson for Graphcore confirmed the setbacks but said the company "remains well-positioned" to capitalize on the AI ​​boom to accelerate commercial adoption opportunities for its chips. Graphcore raised $222 million in its most recent investment round in 2020, valuing it at $2.8 billion at the time. (Xiao Xiao)

The above is the detailed content of Meta has hired a team specializing in AI network chips and is developing an AI supercomputing system. For more information, please follow other related articles on the PHP Chinese website!