Sina technology news May 19, Beijing time morning news, it is reported that Meta Corporation has developed some custom computer chips to help it perform artificial intelligence and video game tasks, the company this week for the first time to communicate with the outside world about these products.
The social media giant first revealed its in-house chip project earlier this week, and then discussed its investment in AI technology infrastructure at a virtual event on Thursday night.
Meta has declared this year an "efficiency year," cutting at least 21,000 jobs and slashing costs. Since then, investors have been closely watching Meta's investments in artificial intelligence and related data center hardware.
While it's expensive for a company to design and develop its own computer chips, Alexis Bjorlin, VP of Infrastructure at Meta, says Meta believes the investment will be justified if it improves the company's performance. The company has also been tweaking the design of its data centers to focus more on energy-efficient technologies, such as liquid cooling and reducing overheating.
One such computer chip, Meta Extensible Video Processor (MSVP), can process video and transmit it to the user while reducing power consumption. Bojolin says that "no commercial product" can process and deliver the 4 billion videos a day as efficiently as Meta aspires to.
The other processor is the first chip in the company's Meta Training and Reasoning Accelerator (MTIA) family, which is designed to help Meta handle a variety of specialized AI tasks. The new MTIA chip deals specifically with "inference," or making predictions or taking actions using AI models that have already been trained.
Bojolin said the new AI reasoning chip helps enhance Meta's recommendation algorithms, which determine what content and ads appear in a user's feed. She declined to answer about the maker of the chip, but a blog post revealed that the processor was "manufactured using Taiwan Semiconductor's 7-nanometer process", suggesting it was contracted by the company.
Bojolin also said Meta has a "multi-generation roadmap" for its AI chip family, including tasks for training AI models, but she declined to provide further details beyond the new reasoning chip. As previously reported, Meta canceled one AI reasoning chip project and started another, but won't be ready for release until 2025. Mr. Bojolin declined to comment.
Because Meta doesn't sell cloud computing services the way Alphabet and Microsoft do, the company doesn't have much incentive to publicly discuss its internal data center chip projects.
"Looking at what we've shared, these are the first two chips we've developed, and it's definitely a glimpse into what we're working on internally." "We don't advertise these things, and we don't need to, but you know, the world is interested," she said.
Aparna Ramani, Meta's vice president of engineering, said the company's new hardware is meant to work well with its homegrown PyTorch software, which has become one of the most common tools used by third-party developers to create AI applications.
The new hardware could eventually be used to perform metaverse related tasks, such as virtual and augmented reality, as well as emerging generative artificial intelligence technologies. Generative AI is a catch-all term for artificial intelligence software capable of creating attractive text, images, and videos.
Ramani also said Meta has developed a generative AI programming assistant for the company's programmers to help streamline software development and operations. The new assistant is similar to the GitHub Copilot tool Microsoft launched in 2021 with the help of AI startup OpenAI.
In addition, Meta also announced that the company's supercomputer Research SuperCluster (" RSC ") has entered its second and final phase. Meta used the supercomputer, which contained 16,000 Nvidia A100 Gpus, to train its LLaMA large language model and other techniques.
Ramani says Meta remains committed to advancing the tech sector 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 is trained using 1.4 trillion tokens. The token refers to the data used to train the AI.
Companies such as OpenAI and Google have not publicly disclosed similar metrics for their large language models, but this week it was revealed that Google's PaLM 2 model uses 3.6 trillion tokens and contains 340 billion parameters.
Unlike other tech companies, Meta released its LLaMA large language model to researchers so they could study the technology. But the LLaMA Grand Language model was subsequently leaked to more people, leading many developers to incorporate the technology into their applications.
Ramani said Meta is "still considering all of our open source collaborations, and of course, I want to reiterate that our philosophy is still open science and cross-collaboration."
Responsible Editor: Zheng Zhuo
User comments