- Meta unveils next-gen custom AI chip, MTIA, optimized for tasks like ad ranking on its platforms.
- The chip, built on a 5nm process, promises up to 3x better performance than its predecessor, MTIA v1.
- Meta aims to achieve greater efficiency and reduce costs by moving towards in-house hardware for AI tasks, though it still relies on GPUs.
Meta introduces MTIA, its latest custom AI chip, promising improved performance. Move signals shift towards in-house hardware for AI tasks while still leaning on GPUs.
Meta’s latest advancement: Unveiling MTIA chip
Meta, determined to rival others in generative AI, invests billions in AI. Funds are allocated for AI researcher recruitment and hardware development, particularly custom chips for AI tasks.
Also read: Intel unveils latest AI chip as competition with Nvidia heats up
Today, Meta unveiled its latest chip, MTIA, a successor to MTIA v1, optimized for tasks like ad ranking on Meta platforms. The new chip, built on a 5nm process, boasts more cores, power, memory, and speed compared to its predecessor.
Enhanced performance and efficiency with MTIA
Live in 16 data center regions, the next-gen MTIA promises up to 3x better performance than its predecessor, though specifics remain vague. Meta claims efficiency gains due to controlling the entire hardware and software stack.
Meta’s transition: From GPU reliance to in-house hardware
The reveal, following a recent generative AI briefing, marks Meta’s shift towards in-house hardware. While not yet utilized for generative AI training, it complements GPUs.
Facing pressure to reduce costs, Meta’s slow hardware progress contrasts with competitors like Google, Amazon, and Microsoft, who are already advanced in custom AI chips. Despite a relatively short development cycle, Meta still trails in achieving GPU independence and matching competitors.
