谷歌于周一正式推出两款全新的专用张量处理单元(TPU)变体——TPU 8tx和TPU 8i Max 1。这两款新品均隶属于今年4月发布的第八代芯片系列 1。
在技术细节方面,TPU 8tx是与AMD合作设计的成果,专注于为机密推理和模型训练提供硬件层的安全隔离 1。而TPU 8i Max则采用了创新封装方式,将两颗TPU 8i芯片合并为一个整体,从而实现内存容量、处理能力以及带宽的翻倍提升 1。
这一举措发生在全球科技巨头竞相开发自研AI半导体以减少对英伟达依赖的大背景下 1。微软、亚马逊以及OpenAI和Anthropic等公司目前也在积极布局此类特定用途的芯片技术 1。
Google has announced the launch of two new variants of its specialized Tensor Processing Units (TPUs), aimed at broadening its artificial intelligence hardware portfolio and reducing reliance on third-party components 1. The newly unveiled chips, named TPU 8tx and TPU 8i Max, were introduced on Monday as part of the eighth-generation chip series originally released in April of this year 1. This move positions Google alongside competitors such as Microsoft, Amazon, OpenAI, and Anthropic, who are also developing proprietary AI semiconductors to decrease their dependence on Nvidia 1.
The TPU 8tx is designed through a collaboration with AMD and focuses specifically on confidential inference and model training, providing security isolation at the hardware layer 1. Meanwhile, the TPU 8i Max enhances performance by combining two TPU 8i chips into a single package, which effectively doubles memory capacity, processing power, and bandwidth 1.
评论
还没有评论,欢迎留下第一条。