显著提升GPU利用率！阿里云AI基础设施成果入选顶会

阿里云

10/16

近日，阿里云提出的GPU池化服务多模型研究成果成功入选顶级学术会议SOSP 2025，该研究提出多模型混合服务系统Aegaeon，可大幅提升GPU资源利用率，目前其核心技术已应用在阿里云百炼平台。SOSP（操作系统原理研讨会）由ACM SIGOPS主办，是计算机系统领域顶级学术会议，平均每年收录的论文数量仅有数十篇，被誉为计算机操作系统界的“奥斯卡”，入选论文代表了操作系统和软件领域最具代表的...

网页链接

免责声明：投资有风险，本文并非投资建议，以上内容不应被视为任何金融产品的购买或出售要约、建议或邀请，作者或其他用户的任何相关讨论、评论或帖子也不应被视为此类内容。本文仅供一般参考，不考虑您的个人投资目标、财务状况或需求。TTM对信息的准确性和完整性不承担任何责任或保证，投资者应自行研究并在投资前寻求专业建议。

热议股票

{"basename":"/hans","ssrTDKData":{"titleTemplate":"%s - 老虎证券","title":"老虎证券全球投资理财平台| 一站式投资美股新股港股A股","description":"老虎证券助您一站式投资美股，新股，港股，A股等全球金融理财产品。新加坡华人最信赖的在线投资平台，现在加入即享低费用，24/5 无时差炒美股投资理财！","keywords":"老虎证券,老虎证券开户,老虎券商,老虎证券官网,老虎证券app,tigertrade老虎证券,股票,炒股,新加坡股票交易平台,投资,投资理财","social":{"ogDescription":"老虎证券助您一站式投资美股，新股，港股，A股等全球金融理财产品。新加坡华人最信赖的在线投资平台，现在加入即享低费用，24/5 无时差炒美股投资理财！","ogImage":"https://c1.itigergrowtha.com/portal5/static/media/og-logo.be62fbe1.png","ogUrl":"https://www.itiger.com/hans/news/2575794782"},"companyName":"老虎证券"},"pageData":{"isMobile":false,"isTiger":false,"isTTM":true,"region":"SGP","license":"TBSG","edition":"fundamental"},"isCrawlerRequest":true,"__swrFallback__":{"@#url:\"https://stock-news.skytigris.cn/v3/news\",params:#id:\"2575794782\",edition:\"fundamental\",auth_exemption:1,,,undefined,":{"share":"https://ttm.financial/m/news/2575794782?lang=zh_CN&edition=fundamental","thumbnail":"","is_english":false,"pubTime":"2025-10-16 14:03","share_image_url":"https://static.laohu8.com/9a95c1376e76363c1401fee7d3717173","id":"2575794782","market":"us","top_or_hot":-1,"title":"显著提升GPU利用率！阿里云AI基础设施成果入选顶会","media":"阿里云","content":"<div>\n<p>近日，阿里云提出的GPU池化服务多模型研究成果成功入选顶级学术会议SOSP 2025，该研究提出多模型混合服务系统Aegaeon，可大幅提升GPU资源利用率，目前其核心技术已应用在阿里云百炼平台。SOSP（操作系统原理研讨会）由ACM SIGOPS主办，是计算机系统领域顶级学术会议，平均每年收录的论文数量仅有数十篇，被誉为计算机操作系统界的“奥斯卡”，入选论文代表了操作系统和软件领域最具代表的...</p>\n\n<a href=\"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20251016141650a6bc7d19&s=b\">网页链接</a>\n\n</div>\n","source":"tencent","html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>显著提升GPU利用率！阿里云AI基础设施成果入选顶会</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n显著提升GPU利用率！阿里云AI基础设施成果入选顶会\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-10-16 14:03 北京时间&nbsp;&nbsp;&nbsp;<a href=http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20251016141650a6bc7d19&s=b><strong>阿里云</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>近日，阿里云提出的GPU池化服务多模型研究成果成功入选顶级学术会议SOSP 2025，该研究提出多模型混合服务系统Aegaeon，可大幅提升GPU资源利用率，目前其核心技术已应用在阿里云百炼平台。SOSP（操作系统原理研讨会）由ACM SIGOPS主办，是计算机系统领域顶级学术会议，平均每年收录的论文数量仅有数十篇，被誉为计算机操作系统界的“奥斯卡”，入选论文代表了操作系统和软件领域最具代表的...</p>\n\n<a href=\"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20251016141650a6bc7d19&s=b\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"LU1003077747.HKD","symbol_name":"BGF GLOBAL EQUITY INCOME \"A6\" (HKDHDG) INC","start_time":0,"source_url":"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20251016141650a6bc7d19&s=b","article_id":"2575794782","we_media_id":null,"thumbnails":[],"rights":{"source":"tencent","url":"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20251016141650a6bc7d19&s=b","rn_cache_url":null,"customStyle":"body{padding-top:10px;}#news_title{font-weight:bold;#titleStyle#;}#news_description span{font-size:12px;#descriptionStyle#;}.footer-note{#statement#}","selectors":".mod-LoadTzbdNews, body","filters":".relate-stock, .hot-list, .recom-box, .wx-sou","directOrigin":true},"url":"https://stock-news.laohu8.com/highlight/detail?id=2575794782","pubTimestamp":1760594588,"columns":[],"sourceInfo":{"source_id":"tencent","name":"腾讯"},"weMediaInfo":null,"summary":"近日，阿里云提出的GPU池化服务多模型研究成果成功入选顶级学术会议SOSP 2025，该研究提出多模型混合服务系统Aegaeon，可大幅提升GPU资源利用率，目前其核心技术已应用在阿里云百炼平台。本届SOSP大会仅收录66篇论文，关于系统软件与AI大模型技术的融合研究成为新的趋势。Aegaeon中对多模型的服务过程主要设计以下三个组：代理层、GPU池与内存管理器。GPU池作为个资源池，包含了组由云提供商虚拟化的GPU实例。","collect":0,"end_time":0,"defaultTopTitle":"qq.com","property":[],"viewcount":null,"language":"zh","relate_stocks":{"LU1003077747.HKD":"BGF GLOBAL EQUITY INCOME \"A6\" (HKDHDG) INC","BK4534":"瑞士信贷持仓","LU0072913022.USD":"UBS (LUX) EQUITY FUND -  GREATER CHINA \"P\" (USD) ACC","LU0880133367.SGD":"UBS (LUX) EQUITY FUND CHINA OPPORTUNITY USD \"P\" (SGD) ACC","LU0553294199.USD":"BGF GLOBAL EQUITY INCOME \"A5G\" (USD) INC","LU1008478684.HKD":"UBS (LUX) EQUITY FUND -  CHINA OPPORTUNITY \"P\" (HKD) ACC","LU1105468828.SGD":"Allianz Total Return Asian Equity AM DIS H2-SGD","LU0661504455.SGD":"Blackrock Global Equity Income A5 SGD-H","BK4558":"双十一","BK4575":"芯片概念","09988":"阿里巴巴-W","LU0918141887.USD":"安联亚洲实际收益股票基金","BK4220":"综合零售","LU0738911758.USD":"Blackrock Global Equity Income A6 USD","LU1235294995.USD":"FIDELITY GLOBAL TECHNOLOGY \"A\" (USDHDG) ACC","LU1769817096.USD":"UBS (LUX) EQUITY SICAV - GLOBAL EMERG MARKETS OPPO \"P\" (USD) INC","BK4524":"宅经济概念","IE00BFMHRM44.USD":"NEUBERGER BERMAN GLOBAL EQUITY MEGATRENDS \"A\" (USD) ACC","LU0106959298.USD":"UBS (LUX) EQUITY FUND - EMERGING MARKETS SUSTAINABLE LEADERS (USD) \"P\" (USD) ACC","LU0821914370.USD":"贝莱德亚洲成长领袖A2","LU1267930227.SGD":"TEMPLETON GLOBAL BALANCED \"AS\" (SGD) ACC A","BABA":"阿里巴巴","BK4588":"碎股","BK4548":"巴美列捷福持仓","LU1046421795.USD":"富达环球科技A-ACC","LU0047713382.USD":"BGF EMERGING MARKETS \"A2\" ACC","BK4565":"NFT概念","BK4538":"云计算","BK4504":"桥水持仓","LU1051769294.HKD":"BGF EMERGING MARKETS EQUITY INCOME \"A6\"(HKDHDG) INC","LU0310800965.SGD":"FTIF - Templeton Global Balanced A Acc SGD","LU1242518931.SGD":"Fullerton Lux Funds - Asia Absolute Alpha A Acc SGD","BK4535":"淡马锡持仓","LU1152091754.HKD":"UBS (LUX) EQUITY FUND - CHINA OPPORTUNITY (USD) \"PM\" (HKD) INC","LU0918141705.HKD":"ALLIANZ TOTAL RETURN ASIAN EQUITY \"AM\" (HKD) INC","LU1048588211.SGD":"Blackrock Asian Dragon A2 SGD-H","BK4554":"元宇宙及AR概念","BK4505":"高瓴资本持仓","BK4581":"高盛持仓","LU1051768304.USD":"贝莱德新兴市场股票收益A6","LU0072462343.USD":"贝莱德亚洲巨龙基金","LU0979878070.USD":"FULLERTON LUX FUNDS - ASIA ABSOLUTE ALPHA \"A\" (USD) ACC","BK4502":"阿里概念","LU1688375341.USD":"贝莱德中国灵活股票基金","LU1823568750.SGD":"Fidelity Global Technology A-ACC SGD","BK4533":"AQR资本管理(全球第二大对冲基金)","LU0797268264.HKD":"ALLIANZ TOTAL RETURN ASIAN EQUITY \"AT\" (HKD) ACC","BK4585":"ETF&股票定投概念","LU1515016050.SGD":"Blackrock Emerging Markets Equity Income A6 SGD-H"},"translate_title":"Significantly improve GPU utilization! Alibaba Cloud AI infrastructure achievements selected for the top conference","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"09988":0.6,"BABA":0.9,"ALBmain":0.6},"content_text":"近日，阿里云提出的GPU池化服务多模型研究成果成功入选顶级学术会议SOSP 2025，该研究提出多模型混合服务系统Aegaeon，可大幅提升GPU资源利用率，目前其核心技术已应用在阿里云百炼平台。SOSP（操作系统原理研讨会）由ACM SIGOPS主办，是计算机系统领域顶级学术会议，平均每年收录的论文数量仅有数十篇，被誉为计算机操作系统界的“奥斯卡”，入选论文代表了操作系统和软件领域最具代表的研究成果。本届SOSP大会仅收录66篇论文，关于系统软件与AI大模型技术的融合研究成为新的趋势。# 研究背景全球模型的数量仍在持续增长，据统计，Hugging Face已托管了超100万个模型。在真实服务场景中，少量热门模型占据了总请求的绝大部分，而超过90%的模型则调用频率较低。当前的事实标准解决方案是为每个模型至少预留一个推理实例，这造成了GPU资源的大量浪费。# 研究方案研究团队在论文中创新性提出多模型混合服务系统Aegaeon，一个实现了token 级多模型混合服务的方法，以实现同时服务众多LLM时提供更高的GPU利用率。Aegaeon中对多模型的服务过程主要设计以下三个组：代理层、GPU池与内存管理器。图｜研究框架图代理层（Proxy layer）。代理层负责接收和分发推理requests，确保负载均衡和容错能⼒，状态同步是通过共享内存机制实现的，例如Redis。Aegeeon可以将不同模型的requests分发到同⼀个实例。GPU池（GPU pool）。GPU池作为⼀个资源池，包含了⼀组由云提供商虚拟化的GPU实例。每个实例可能包含托管在⼀台物理机器上的⼀个或多个GPU。在Aegaeon中，单个实例执⾏预填充（prefill）或解码（decoding）作业，并且可以在token级调度器的指导下，为不同模型的requests提供服务。这使得模型切换成为⼀个关键操作。内存管理器（Memory manager）。内存管理器在服务集群中协调每个节点的主机内存和空闲GPU内存，⽬的有两个：1）QuickLoader利⽤可⽤内存来缓存模型权重，加速从远程模型仓库获取模型的耗时过程；2）GPU-CPU KV管理机制统一存储和管理KV缓存。# 研究创新点与结论Aegaeon多模型混合服务系统，首次将调度实现在token级别，在每次生成完下一个token之后，都可通过精确的执行时间预测和创新的token级调度算法规划是否需要切换模型，从而实现多个模型混合服务且满足延迟要求。同时，通过组件复用、显存精细化管理和KV缓存同步优化等全栈技术，Aegaeon将模型切换开销降低97%，确保了token级调度的实时性，可支持亚秒级的模型切换响应。Aegaeon系统支持单GPU同时服务多达7个不同模型，相比现有主流方案提升1.5-9倍的有效吞吐量，实现2-2.5倍的请求处理能力。图｜Aegaeon显著提升GPU利用率# 成果应用目前，Aegaeon核心技术已在阿里云百炼平台部署，服务数十个模型的推理，将服务这些模型所需的GPU数量减少了82%。截至目前，阿里云百炼平台已上线Qwen、Wan、DeepSeek等200多款业界领先的模型，过去一年，阿里云百炼平台的模型调用量增长了15倍。","kind":"news","is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"1","news_tag":"corporation","news_rank":0,"symbols":[],"gpt_button":0,"need_auth":false,"code":"91000000","status":"200"}}}