Meta Develops Its First Custom AI Chip: Power Consumption Lower than NVIDIA's

On Thursday, Meta announced the introduction of its first custom chip designed for artificial intelligence models. Unlike competitors such as Google and Microsoft, which primarily focus on cloud computing services, Meta had not previously disclosed its internal chip projects. However, growing interest from the tech community has encouraged the company to enhance transparency about its innovations.

The new chip, named the "Meta Training and Inference Accelerator" (MTIA), boasts a power consumption of just 25 watts—significantly lower than offerings from rivals like Nvidia. According to Meta, the initial MTIA chip is optimized for AI inference tasks. Joel Coburn, a software engineer at Meta, stated that while the company initially considered GPUs for inference, they proved to be less efficient in real-world applications and costly to deploy. “This is why we need the MTIA,” Coburn emphasized.

Additionally, Meta introduced the "Scalable Video Processor" (MSVP), engineered for efficient video processing and transmission with reduced energy consumption. The company emphasized the necessity of this technology to handle an impressive 4 billion videos daily, underscoring the importance of efficient video management in today’s digital era.

Most people like

Find AI tools in YBX