如果您希望可以时常见面，欢迎标星收藏哦~ 思科今日发布Silicon One G300，这是一款专为大规模人工智能集群构建而设计的102.4 Tbps交换 芯片，持续推动网络向人工智能创新平台转型。思科Silicon One G300将为全新的思科N9000和思 科8000系统提供动力，进一步拓展数据中心人工智能网络的边界。这些系统采用创新的液冷技 术，并支持高密度光模块，从而达到新的效率标杆，确保客户充分利用其GPU投资。此外，思科 还增强了Nexus One，使企业能够更轻松地运维其人工智能网络（无论是在本地还是在云端），从 而消除可能阻碍企业扩展人工智能数据中心的复杂性。 思科总裁兼首席产品官 Jeetu Patel表示："我们正在通过全栈创新——从芯片到系统和软件——引 领人工智能网络的性能、可管理性和安全性。我们正在为未来的基础设施奠定基础，支持从超大规 模企业到大型企业等各类客户向人工智能驱动的工作负载转型。" "随着人工智能训练和推理规模的不断扩大，数据移动成为高效人工智能计算的关键；网络本身也 成为计算的一部分。这不仅仅关乎更快的GPU——网络必须提供可扩展的带宽以及可靠、无拥塞 的数据传输， ...

Core Insights - Cisco has introduced the Silicon One G300, a 102.4 Tbps switching silicon aimed at enhancing AI networking capabilities in data centers [3][5] - The new Cisco N9000 and Cisco 8000 systems, powered by the G300, are designed to support massive AI cluster buildouts with improved efficiency and reliability [3][10] - Innovations include liquid cooling and high-density optics, which contribute to a nearly 70% improvement in energy efficiency [7][14] Product Features - The Silicon One G300 features Intelligent Collective Networking, which enhances network utilization by 33% and reduces job completion time by 28% compared to non-optimized path selection [6] - The new systems support gigawatt-scale AI clusters and are designed for various customer types, including hyperscalers and enterprises [7][10] - Cisco's Nexus One provides a unified management plane, simplifying operations and enabling faster deployment of AI networks [11][15] Performance and Efficiency - The G300's architecture allows for high-performance, programmable networking, which is crucial for maximizing GPU utilization and ensuring reliable data movement [4][7] - The introduction of 1.6T OSFP optics and 800G Linear Pluggable Optics (LPO) significantly enhances bandwidth connectivity while reducing power consumption by 50% [14] - The liquid-cooled systems achieve the same bandwidth previously requiring six older systems, showcasing a leap in performance and efficiency [14] Ecosystem and Partnerships - Cisco collaborates with key technology partners like AMD, Intel, and NVIDIA to deliver optimized infrastructure for AI workloads [16][18] - The integration of Cisco's networking solutions with partners' technologies aims to enhance performance and interoperability across diverse environments [17][19] - Cisco's approach addresses the growing demand for scalable, energy-efficient AI infrastructure as adoption expands beyond hyperscalers [18][19]