Santosh Janardhan, Meta’s VP and Head of Infrastructure, shared details of the company’s progress in a recent blog post, emphasizing the growing interest in AI within Meta. The new chip, named MTIA (Meta Training and Inference Accelerator), is a custom accelerator chip family tailored for inference workloads. By combining MTIA chips with GPUs, Meta aims to deliver enhanced performance, reduced latency, and increased efficiency for its workloads.
The development of MTIA chips is a significant move for Meta, as it allows the company to compete with other tech giants like Google and Microsoft, who have already made substantial investments in AI technologies. Moreover, as the chip is designed in-house, it can be optimized to precisely match Meta’s specific workload requirements, freeing the company from relying on external providers.
