**Meta 发布新一代AI芯片，性能提升三倍，瞄准生成式人工智能**
Meta Platforms 近日宣布推出其训练与推理加速器项目（MTIA）的最新迭代，这一专为AI工作负载设计的定制芯片系列展现了显著的性能提升。新一代MTIA芯片旨在提供计算、内存带宽和内存容量之间的理想平衡，以优化AI处理效率。
据透露，这款新芯片拥有256MB的片上内存，主频达到1.35GHz，采用先进的台积电5纳米工艺制造。相比第一代产品，内存容量翻了一番，主频提高了约69%，性能整体提升了三倍。这些改进将极大地增强Meta的AI基础设施，特别是在处理排名和推荐算法方面。
Meta的长远目标不仅限于当前的应用，该公司还表示，计划最终将MTIA芯片的功能扩展至训练生成式人工智能，如其自家的Llama语言模型。这一战略调整意味着Meta正积极布局AI前沿技术，以支持更复杂、更智能的交互和内容生成。
这一发布凸显了Meta在AI硬件领域的持续投入，旨在保持其在全球社交媒体和数字通信领域的领先地位。通过不断提升硬件性能，Meta能够更有效地处理海量数据，为用户提供更个性化、更智能的体验，同时也为未来的AI创新奠定了坚实的基础。
英语如下：
**News Title:** “Meta Unveils New AI Chip: Doubled Performance Targets Generative AI”
**Keywords:** Meta AI Chip, Performance Boost, MTIA Launch
**News Content:**
Meta Platforms has recently announced the latest iteration of its Machine Learning Training and Inference Accelerator (MTIA) project, a custom chip series designed specifically for AI workloads, showcasing substantial performance enhancements. The new MTIA chip aims to provide an ideal balance between compute, memory bandwidth, and memory capacity to optimize AI processing efficiency.
The chip boasts 256MB of on-chip memory, a clock speed of 1.35GHz, and is fabricated using TSMC’s advanced 5-nanometer process. Compared to its first-generation counterpart, it has double the memory capacity, a roughly 69% increase in clock speed, and an overall tripling of performance. These enhancements are set to significantly bolster Meta’s AI infrastructure, particularly in the handling of ranking and recommendation algorithms.
Beyond current applications, Meta has disclosed plans to eventually expand the MTIA chip’s capabilities to include training generative AI models, such as its in-house Llama language model. This strategic move signals the company’s proactive positioning in the forefront of AI technology to support more complex and intelligent interactions and content generation.
This launch underscores Meta’s ongoing investment in AI hardware, striving to maintain its leadership in global social media and digital communication. By continually improving hardware performance, Meta can more effectively handle vast amounts of data, delivering more personalized and intelligent experiences to users while laying a solid foundation for future AI innovations.
【来源】https://www.ithome.com/0/761/105.htm
Views: 1