Workflow
华为910B
icon
Search documents
风电助力陆数海算 上海临港探索算电协同新范式
Sou Hu Cai Jing· 2025-08-12 13:47
Core Insights - The article discusses the increasing demand for computing power driven by the AI boom and the corresponding rise in electricity consumption, emphasizing the need for more green energy solutions to reduce costs and improve efficiency [1] Group 1: Project Overview - The world's first "land-sea computing + wind energy integration" project is being deployed at the Shanghai Lingang International Data Port, featuring a unique four-story underwater data center [1][3] - The project, undertaken by Hailan Cloud Technology Co., Ltd., has a total investment of 1.6 billion yuan and a total scale of 24 megawatts, with a green electricity supply rate exceeding 90% [5][6] Group 2: Technological Innovations - The underwater data center will house 192 cabinets, each with a capacity of 12 kilowatts, directly connected to offshore wind power [3] - The project aims to utilize seawater for natural cooling to reduce energy consumption, a technology previously tested only by Microsoft in 2015 [6] Group 3: Power Transfer and Efficiency - On July 8, China Telecom's Lingang computing center successfully transferred AI computing tasks over 1,000 kilometers to a data center in Hubei, demonstrating the feasibility of real-time "East Data West Computing" [7] - This transfer allows for quick switching of computing power to areas with lower electricity prices, maximizing resource efficiency [9] Group 4: Future Developments - The data center is designed to withstand extreme weather conditions, with plans to deploy offshore wind power further into the sea as costs for offshore wind energy have dropped below 0.3 yuan per kilowatt-hour [8] - The industry calls for the establishment of technical standards and a national-level "computing power exchange" to facilitate the development of a collaborative computing and electricity model [11]
英伟达H20不让用?全国产算力深度推理模型讯飞星火X1升级,4张华为910B即可部署满血版
量子位· 2025-04-21 13:23
就在今天, 最新国产深度推理大模型升级更新 ,算力构成全国产化,在模型参数比业界同类模型小一个数量级的情况下,实现了整体效果对 标业界最高水平。 这就是讯飞星火,最新升级的推理模型X1。 一方面是通用任务效果显著提升,在通用任务效果评测中全面对标OpenAI o1和DeepSeek R1,尤其在数学、知识问答等方面表现突出。 (测试集合来源:自建测试集主要来自真实的大模型请求任务数据,来源分布包括讯飞星火APP、星火大模型API、业界主流任务数据等;公开测试集主要以数 学、答题、推理、代码等外部典型测试集为主。) 另一面值得关注的,它还 是 业界唯一一个基于全国产算力训练的推理模型。 在波云诡谲的当下,尤其算力新一波打压H20被禁之后,这样的 模型展现出来的效果和实力显得尤为受到关注。 白交 发自 凹非寺 量子位 | 公众号 QbitAI 英伟达H20也不能用了。中国大模型还能好吗? 能 。 此外,升级后的星火X1私有化部署简便,定制门槛低—— 4张910B即可部署满血版星火X1,16张910B即可完成行业定制优化 ,再次印证全栈自主可控大模型,具备登顶高水平的实力和持续创新的 潜力。 星火X1升级:长思维链 ...