上海的陆家嘴

Meta Platforms 近日宣布推出了其训练与推理加速器项目(MTIA)的最新迭代产品,这款新一代AI芯片的性能是初代的三倍,旨在优化和加速AI工作负载。MTIA芯片系列是Meta专门为人工智能设计的定制化解决方案,目前主要用于支持排名和推荐算法的训练。

据悉,Meta的长远目标是扩展该芯片的功能,使其能够处理更为复杂的任务,例如训练生成式人工智能,如公司自家的Llama语言模型。新款MTIA芯片在设计上注重计算能力、内存带宽和内存容量之间的平衡,以实现更高效的AI运算。该芯片配备了256MB的片上内存,主频达到1.35GHz,这一配置相比第一代的128MB内存和800MHz主频有了显著提升。

这款芯片采用台积电先进的5纳米工艺制造,技术上的进步使得芯片在性能增强的同时,也能够保持能效。Meta的这一举措彰显了其在AI硬件领域的持续投入和创新,旨在为用户提供更加智能化和个性化的服务体验。随着AI技术的不断发展,Meta的MTIA芯片有望在未来的AI应用中发挥关键作用。

英语如下:

News Title: “Meta Unveils New AI Chip: Doubled Performance, Targeting Generative AI”

Keywords: Meta MTIA, Chip Upgrade, AI Accelerator

News Content: Meta Platforms recently announced the latest iteration of its Meta Training and Inference Accelerator (MTIA) project, a new AI chip with triple the performance of its predecessor, designed to optimize and speed up AI workloads. The MTIA chip series is a customized solution tailored by Meta specifically for artificial intelligence, currently主要用于 powering ranking and recommendation algorithm training.

It is reported that Meta’s long-term objective is to expand the chip’s capabilities to handle more complex tasks, such as training generative AI models like the company’s in-house Llama language model. The new MTIA chip emphasizes a balance between computational power, memory bandwidth, and memory capacity for more efficient AI operations. It boasts 256MB of on-chip memory and a clock speed of 1.35GHz, a significant upgrade from the first generation’s 128MB memory and 800MHz clock speed.

Manufactured using TSMC’s advanced 5nm process, the technological advancements allow the chip to enhance performance while maintaining energy efficiency. Meta’s move underscores its ongoing investment and innovation in AI hardware, aiming to provide users with more intelligent and personalized service experiences. As AI technology continues to evolve, Meta’s MTIA chips are poised to play a crucial role in future AI applications.

【来源】https://www.ithome.com/0/761/105.htm

Views: 1

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注