Shulou(Shulou.com)11/24 Report--

Thanks to the CTOnews.com netizen Mr. Air, the left hand of Brother Black fly, and the delivery of the clue of Xiao Zhanche! According to news in the morning of May 19, Beijing time, Meta reportedly developed some customized computer chips to help it perform artificial intelligence and video games tasks, and the company communicated with the outside world for the first time this week.

The social media giant first unveiled its internal chip project earlier this week and then discussed its investment in artificial intelligence technology infrastructure at a virtual event on Thursday night.

Meta had previously declared this year a "year of efficiency", cutting at least 21000 jobs and slashing costs. Since then, investors have been keeping a close eye on Meta's investment in artificial intelligence and related data center hardware.

While it is expensive for a company to design and develop its own computer chips, Alexis Bjorlin, vice president of infrastructure at Meta, says Meta believes the investment makes sense as long as it improves the company's performance. The company has also been adjusting the design of its data centers to focus more on energy-efficient technologies, such as liquid cooling and reducing overheating.

One of the computer chips, called Meta Extensible Video processor (MSVP), can process video and transmit it to users, while reducing energy consumption. Mr Bojolin says there is "no commercial product" that can process and transmit 4 billion videos a day as efficiently as Meta craves.

The other processor, the first chip in the company's family of Meta training and reasoning Accelerator (MTIA), is designed to help Meta handle a variety of specialized artificial intelligence tasks. The new MTIA chip specializes in "reasoning", which uses trained artificial intelligence models to make predictions or take actions.

Bojolin says the new AI inference chip will help enhance Meta's recommendation algorithm, which determines what content and ads are displayed in a user's information stream. She declined to comment on the maker of the chip, but a blog post revealed that the processor was "manufactured using TSMC's 7nm process", indicating that the processor was manufactured by TSMC.

Ms Bojolin also said Meta had developed a "multi-generation roadmap" for her AI chip family, including tasks to train AI models, but declined to provide further details beyond the new reasoning chips. It has been reported that Meta has cancelled one artificial intelligence reasoning chip project and launched another, but will not launch a specific product until 2025. Mr Bo Qiaolin declined to comment.

Since Meta does not sell cloud computing services like Alphabet and Microsoft, the company has little incentive to discuss its internal data center chip project publicly.

"if you look at what we share, these are the first two chips we developed, so we can definitely get a glimpse of the projects we are working on internally." "We don't promote these things, and we don't need to promote them, but you know, the whole world is interested," she said. "

Aparna Ramani, vice president of engineering at Meta, said the company's new hardware was designed to work efficiently with its homegrown PyTorch software, which has become one of the most commonly used tools for third-party developers to develop artificial intelligence applications.

The new hardware could eventually be used to perform meta-universe-related tasks, such as virtual reality and augmented reality, as well as emerging generative artificial intelligence technologies. Generative artificial intelligence is a general term, which refers to artificial intelligence software that can create attractive text, images and videos.

Ramani also said that Meta has developed a generative artificial intelligence programming assistant for the company's programmers to help them simplify the software development and operation process. The new assistant is similar to the GitHub Copilot tool that Microsoft launched in 2021 with the help of artificial intelligence startup OpenAI.

In addition, Meta also announced that the company's supercomputer Research SuperCluster ("RSC") has entered the second and final phase. This supercomputer contains 16000 Nvidia A100 GPU,Meta using this supercomputer to train its LLaMA big language model and other techniques.

Ramani said that Meta still adheres to its belief and hopes to promote the development of science and technology by contributing to open source technology and artificial intelligence research. The company revealed that its largest LLaMA large language model, LLaMA 65B, contains 65 billion parameters and uses 1.4 trillion tokens for training. Tokens refer to the data used to train artificial intelligence.

Companies such as OpenAI and Google have not publicly disclosed similar indicators for their big language models, but media revealed this week that Google's PaLM 2 model uses 3.6 trillion tokens and contains 340 billion parameters.

Unlike other technology companies, Meta has released its LLaMA large language model to researchers so that they can study the technology. But the LLaMA big language model was then leaked to more people, causing many developers to integrate the technology into their applications.

Meta "is still considering all our open source partnerships, and of course, I would like to reiterate that our philosophy is still open science and cross-collaboration," Mr Ramani said.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.