深圳首例昇腾384超节点算力集群落地商用
2 1 Shi Ji Jing Ji Bao Dao·2025-12-01 14:36

Core Insights - The launch of the Ascend 384 super node computing cluster in Shenzhen marks a significant advancement in AI computing capabilities, transitioning from traditional CPU-based systems to a more integrated and efficient architecture [1][2]. Group 1: Project Overview - The Ascend 384 super node is a collaborative project between Huawei and the Shenzhen Longgang District Urban Investment Group, designed to enhance AI computing power [1]. - This project represents a shift in AI computing clusters from "patchwork integration" to "system fusion," addressing issues like high communication latency and low collaboration efficiency [1]. Group 2: Technical Advancements - The new cluster features 384 high-performance NPUs and integrates 192 of the latest high-performance Kunpeng CPUs, achieving a threefold increase in training performance and a fourfold increase in inference performance [1][2]. - The architecture allows any processor within the cluster to access data center resources as if they were local memory, significantly improving overall system performance [1]. Group 3: Strategic Implications - The deployment of the Ascend 384 super node is a key initiative in the Longgang District's strategy to develop a comprehensive AI resource pool, aiming to provide robust support across various sectors such as government, finance, energy, and manufacturing [2]. - The project aligns with Longgang's "All in AI" strategy, which includes initiatives like the CANN open-source strategy and the establishment of a data company to leverage extensive data resources [2]. Group 4: Ecosystem Development - The collaboration between CANN as the AI computing engine and the open-source HarmonyOS as the operating system aims to create a seamless flow of computing resources across multiple devices [3]. - This integrated approach is expected to attract global AI algorithm developers and HarmonyOS application developers to Shenzhen, fostering a collaborative technology ecosystem and accelerating the development of the AI industry in the Greater Bay Area [3].

深圳首例昇腾384超节点算力集群落地商用 - Reportify