Core Insights - Amazon Web Services (AWS) has launched its next-generation in-house server CPU, Graviton5, which is claimed to be the most powerful and energy-efficient data center CPU to date [1][2] - The introduction of Graviton5 and the AI ASIC chip Trainium3 highlights AWS's ongoing expansion of its chip product lineup and a shift towards in-house development [1][2] - AWS's self-developed Graviton series CPUs now account for over half of the new large-scale CPU clusters added in the past three years, indicating a significant reduction in reliance on Intel and AMD [1] Product Details - Graviton5 offers up to a 25% performance improvement compared to its predecessor while maintaining industry-leading energy efficiency, enabling faster application performance and reduced computing costs [2] - The L3 cache capacity of each Graviton5 core is 2.6 times that of Graviton4, with network and storage bandwidth increased by 15% to 20% [2] - The M9g instance based on Graviton5 is currently in preview, while C9g and R9g instances are planned for release in 2026 [3] Market Context - The trend of cloud giants like Amazon, Microsoft, and Google developing their own server CPUs and AI accelerators may significantly reduce their procurement of server CPUs from Intel and AMD, as well as the overall demand for NVIDIA's AI GPU clusters [1][5] - The shift towards in-house ARM architecture CPUs allows cloud providers to optimize for specific workloads, potentially leading to significant cost savings and improved performance [5][6] - Despite the rise of ARM-based CPUs, x86 architecture (Intel and AMD) remains the dominant choice in the global server market, particularly for traditional enterprise cloud infrastructure [6]
性能提升25%!亚马逊新一代自研服务器芯片Graviton5问世 冲击x86霸权