Meta Platforms Inc (NASDAQ:FB) has not only created an artificial intelligence (AI) technology, called LLaMA, that can power online chatbots, but the Silicon Valley giant has now unveiled its newly built custom computer chips to help with its AI and video-processing tasks.
CNBC reported that Meta talked about its in-house silicon chip projects for the first time to reporters this week, ahead of a virtual event on Thursday while discussing its AI infrastructure investments.
It is both expensive and unheard of for a social networking company to design and build its own computer chips, but Meta's vice president of infrastructure Alexis Bjorlin told CNBC that Meta believes that the “improved performance will justify the investment.”
Bjorlin told CNBC that “there was nothing commercially available” that could handle the task of processing and delivering four billion videos a day as efficiently as Meta wanted.
As a result, one of the newly minted computer chips, the Meta Scalable Video Processor, or MSVP, is used to process and transmit video to users, while cutting down on energy requirements.
The other processor, the Meta Training and Inference Accelerator, or the MTIA family of chips is aimed at helping with AI-specific tasks.
Meanwhile, Meta's vice president of engineering Aparna Ramani, said the tech giant continues to act on its belief that it should contribute to open-source technologies and AI research to advance the technology.
Open science and collaboration
In February, Meta took the industry by surprise by releasing its AI system’s underlying computer code. It disclosed that its biggest LLaMA language model, LLaMA 65B, contains 65 billion parameters and was trained on 1.4 trillion tokens, which refers to the data used for AI training.
The New York Times noted that Meta CEO Mark Zuckerberg’s highly “unusual move” in the rapidly evolving world of AI was tantamount to literally giving “away its AI crown jewels.”
Academics, government researchers and others who give their email address to Meta can download the code once Meta vets the individual, according to the NYT.
Ramani said Meta is “still thinking through all of our open source collaborations, and certainly, I want to reiterate that our philosophy is still open science and cross collaboration.”
In addition, Ramani said Meta’s new hardware was developed to work with its home-grown PyTorch software, which has become popular with third-party developers to create AI apps.
The new hardware will eventually be used to power metaverse-related tasks, such as virtual reality and augmented reality, as well as generative AI which creates text, images, and videos.
Interestingly, Meta has created a generative AI-powered coding assistant for its own developers to help them create and operate software. The new assistant is similar to Microsoft’s GitHub Copilot tool, released in 2021 with help from the AI startup OpenAI.
In addition, Meta said it completed the second-phase, or final, buildout of its supercomputer dubbed Research SuperCluster, or RSC.
Contact the author Uttara Choudhury at uttara@proactiveinvestors.com
Follow her on Twitter: @UttaraProactive