登录
立即咨询

语言选择

谐云 谐云
在这里探索云原生
云原生:AI的“神队友”,让算力不再“卡脖子”
2025年04月02日

2025年,全球AI产业正经历一场空前的“算力饥渴症”。GPT-5 训练消耗的电力足以支撑一座百万人口城市运行一周,单次训练成本突破 5亿美元; Grok 3 的训练并非如此前传闻的在“10 万张 GPU 上进行”,而是使用了“20 万张 GPU”。对此,有网友指出其算力消耗是 DeepSeek V3 的 263 倍; 中国 DeepSeek-V3 训练数据量相当于GPT3的49倍。模型训练所使用的芯片主要是2000多块的NVIDIA H800 GPU组成的集群。

大模型密度定律仍未被被打破,伴随模型能力密度的指数级增强,算力需求会持续扩张。 

数据来源:浙大deepseek公开课第一季第二期;智云研报《2025年DeepSeek模型优势:算力、成本角度解读报告》;峰哥Python笔记《20万张GPU集群如何训练AI模型:DeepSeek、Grok3、GPT-4的成本、电力及基础设施解析》

 

何为算力?

算力并非是一个新兴的概念。从公元前就开始出现了依赖人工操作的非智能计算辅助工具,例如手指、结绳、算筹、算盘等。到达17世纪,人类进入了机械计算时代。帕斯卡计算器是1642年出现的首个机械加法器,它可以通过齿轮传动实现简单运算。

20世纪,人类是实现了从电子管计算机到超级计算机的飞速发展。计算机进入商用和科研领域,并可以为天气预报、核武器模拟等复杂科学计算提供支持。21世纪现代算力实现多维扩展,从分布式计算与云计算、 异构计算与专用芯片(CPU,TPU),到边缘计算与物联网。算力从人工到机械、电子、分布式再到智能化的演变,本质是人类突破自身生理与物理限制的过程。

算力与人工智能—"燃料"与"引擎"

AI性能 ≈ 算法效率 × 数据质量 × 算力规模

其中,算力是唯一可被量化的“硬指标”——它直接决定了模型训练的速度、规模与可能性。

算力的“摩尔定律困境”: 2012年,训练AlexNet(CNN鼻祖)仅需2块GPU,耗时5天;而2023年训练GPT-4需数万块A100芯片,耗电超50GWh(相当于一个小型城市年用电量)。算力需求每3-4个月翻一番,远超芯片制程迭代速度。

                                                           大模型所需算力变化

对中国的算力封锁

2022年10月,禁止A100/H100等高端GPU对华出口。

2023年10月,限制英伟达特供版A800/H800、AMD MI250,并将制裁延伸至云算力服务。

光刻机封锁,荷兰ASML被禁止向中国出售EUV及部分DUV光刻机,中芯国际14nm以下先进制程量产受阻。

芯片制造设备,应用材料、泛林等企业被限制对华出口刻蚀机、沉积设备。

 

国内外主流算力能力对比

目前,AI大模型的竞争本质是算力的军备竞赛。然而国产芯片在工艺和生态上仍处劣势。

1.性能差距:单卡算力与互联技术落后

国外巨头:英伟达H100(FP16算力 2000 TFLOPS)、AMD MI300X(显存 192GB)等芯片,配合 NVLink高速互联技术(带宽900GB/s),可构建超大规模计算集群。

国内现状:华为昇腾910B(FP16算力 320 TFLOPS)、寒武纪思元590(显存64GB)等国产芯片,性能约为H100的15%-30%;互联技术多依赖PCIe 4.0(带宽64GB/s),集群效率仅为英伟达方案的30%-50%。

2.成本差距:单位算力价格翻倍

英伟达H100单卡售价约 3万美元,而国产芯片因制程(如14nm vs 台积电4nm)和生态劣势,单位算力成本高出2-3倍。若训练GPT-3级别模型,国内企业需多投入数千万美元。

云原生冲破算力封锁,AI时代的“弹性救星”

从前文的AI 模型训练的算力用量趋势图我们可以发现,算力需求呈现动态上升趋势。这说明算力并非是决定大模型成功与否的唯一标准。对现有算力调度提升与挖掘,是渡过目前中国算力封锁的重要方法,穷则战术穿插。

当芯片禁令撞上AI算力迫切需求,一场由代码驱动的算力解放运动正在中国云计算战场上悄然崛起。从DeepSeek首创动态计算流引擎(DCFE),提出5D混合并行框架等就证明了通过算法与生态将算力从物理约束中解放,为国产AI生态开辟了一条“以架构创新换制程差距”的突围路径的可行性和有效性。

而这种“以软制硬”的生态革命手段正是云原生的强项。云原生技术正以破局者的姿态,将"弹性基因"注入计算体系的毛细血管。云原生技术体系通过容器化、微服务编排和动态资源调度机制,为AI算力瓶颈问题提供了系统性的解决方案。

核心价值在于通过以下技术路径实现算力资源的高效利用:

1.基于Kubernetes的智能调度算法

可实时分析集群负载状态,在CPU/GPU异构计算环境中自动匹配最优任务分配策略,将AI训练任务分解至可用计算节点,使硬件资源利用率提升40%-65%。

Volcano调度器优化LLM训练

字节跳动提出强化学习LLM Agent框架AGILE。在Kubernetes上扩展Volcano调度器,引入GANG调度算法解决大模型训练任务碎片化问题,开发DRF(Dominant Resource Fairness)策略实现GPU显存与计算单元的双维度调度。通过拓扑感知调度自动构建AllReduce通信环,将3500张A100显卡组成的集群利用率从58%提升至89%。

2.容器化混合部署

容器化的资源隔离特性支持混合部署场景,允许不同优先级、不同架构的AI工作负载共享物理资源,显著降低因资源碎片化导致的算力浪费。

NVIDIA DGX SuperPOD的异构架构混合部署

统一GPU池化:通过NVIDIA vGPU Manager将A100显卡分割为7个vGPU实例 架构无关封装:使用Singularity容器封装不同指令集的应用程序。DGX SuperPOD 混合部署可实现 80%+ GPU 利用率(传统场景通常低于 50%)。

来源:NVIDIA 官方发布白皮书《Maximizing GPU Utilization with DGX SuperPOD》

阿里巴巴双11弹性资源池

双11期间流量峰值达日常的300倍,传统部署需提前数月扩容服务器,资源浪费率超50%。阿里巴巴双11采用容器化混合部署,基于Kubernetes(ACK)和自研Koordinator调度器,实现高优先级在线交易与低优先级离线任务(如数据分析)的动态资源调度。混部后集群CPU利用率从18%提升至75%,兼顾高弹性、低成本与业务零降级稳定性。(https://xie.infoq.cn/article/c2ed4baeeb632321766c5e912)

3.弹性伸缩机制结合细粒度监控指标

(如GPU显存占用率、张量计算单元利用率)可在毫秒级时间窗口内动态调整计算资源配置,有效应对AI模型训练中突发的大规模并行计算需求。

阿里云

LLM推理场景的潮汐流量特征,阿里云ACK集群结合Knative和Fluid技术实现。动态资源抢占:基于GPU利用率(监控粒度达毫秒级)和QPS指标,低峰时段释放50%GPU资源用于离线训练,高峰时段通过优先级策略抢占资源,整体资源闲置率下降60%。(来源:公众号阿里云基础设施)

4.资源池化效应

云原生架构的标准化接口设计打破了传统云计算平台的技术锁定,支持跨公有云、私有云及边缘节点的统一算力调度,形成分布式AI训练的资源池化效应。这些技术特性共同构建了面向AI工作负载的弹性计算基础设施,在硬件资源总量受限的情况下,通过提升单位算力产出效率来突破计算瓶颈。

多云资源统一纳管

火山引擎的分布式云原生平台(DCP)通过统一资源池化管理,支持跨AWS、阿里云、华为云等平台的GPU资源动态调度。其核心策略包括:智能分发:根据GPU型号、区域库存、成本优先级自动匹配最优资源;故障重调度:在单云资源不足时,自动切换至其他云或边缘节点;弹性伸缩:常驻资源与突发扩容资源的混合编排,降低闲置成本。

算力资源池化技术

趋动科技的猎户座OrionX通过GPU虚拟化技术,将物理GPU资源拆分为细粒度虚拟单元,实现跨服务器的共享调度。华为CloudMatrix进一步将CPU、NPU、内存等资源全互联池化,形成紧耦合的多元算力池,支持AI任务按需组合资源。此类技术可将GPU利用率从传统独占模式的30%提升至80%以上。 

 

谐云算力管理平台  

助力企业算力应用与管理

云原生的价值,在于用“软实力”弥补“硬差距”。

通过弹性调度、异构融合与算法创新,想象云原生技术能将国产芯片编织为全球效率顶尖的算力网络。通过跨平台资源池化、智能编排与软硬协同优化,云原生让每一焦耳能量迸发出超越物理硬件的智能密度。 

 

添加评论
谐云
2025年04月19日
I'm really impressed with your writing skills and also with the layout on your blog. Is this a paid theme or did you customize it yourself? Either way keep up the nice quality writing, it抯 rare to see a nice blog like this one these days..
添加回复
谐云
2025年04月19日
添加回复
回复:Hi, i think that i saw you visited my site thus i came to “return the favor”.I'm trying to find things to enhance my site!I suppose its ok to use some of your ideas!!
添加回复
回复:This information is invaluable. Where can I find out more?
添加回复
回复:Wonderful blog! Do you have any suggestions for aspiring writers? I'm hoping to start my own website soon but I'm a little lost on everything. Would you suggest starting with a free platform like Wordpress or go for a paid option? There are so many choices out there that I'm completely confused .. Any ideas? Many thanks!
添加回复
回复:Thanks on your marvelous posting! I quite enjoyed reading it, you are a great author. I will always bookmark your blog and will come back very soon. I want to encourage you to definitely continue your great posts, have a nice weekend!
添加回复
回复:I'm impressed, I have to admit. Rarely do I encounter a blog that's both educative and entertaining, and let me tell you, you have hit the nail on the head. The issue is an issue that too few folks are speaking intelligently about. Now i'm very happy that I came across this in my hunt for something relating to this.
添加回复
回复:Having read this I believed it was very enlightening. I appreciate you taking the time and energy to put this article together. I once again find myself spending a significant amount of time both reading and posting comments. But so what, it was still worthwhile!
添加回复
回复:Write more, thats all I have to say. Literally, it seems as though you relied on the video to make your point. You definitely know what youre talking about, why throw away your intelligence on just posting videos to your blog when you could be giving us something informative to read?
添加回复
回复:Every weekend i used to pay a quick visit this site, because i want enjoyment, for the reason that this this web page conations really good funny information too.
添加回复
回复:Terrific article! This is the kind of info that are meant to be shared across the web. Shame on Google for now not positioning this publish higher! Come on over and talk over with my website . Thanks =)
添加回复
回复:This website was... how do I say it? Relevant!! Finally I have found something which helped me. Kudos!
添加回复
回复:Its not my first time to visit this website, i am visiting this site dailly and take nice data from here everyday.
添加回复
回复:Hi there every one, here every person is sharing such know-how, so it's good to read this website, and I used to pay a visit this website everyday.
添加回复
回复:fantastic points altogether, you simply received a brand new reader. What may you recommend about your put up that you simply made some days ago? Any positive?
添加回复
回复:I am really loving the theme/design of your site. Do you ever run into any internet browser compatibility issues? A handful of my blog readers have complained about my website not working correctly in Explorer but looks great in Opera. Do you have any suggestions to help fix this problem?
添加回复
回复:If you wish for to get a great deal from this post then you have to apply these methods to your won website.
添加回复
回复:Hi there! This is my 1st comment here so I just wanted to give a quick shout out and tell you I genuinely enjoy reading your articles. Can you suggest any other blogs/websites/forums that go over the same topics? Thanks a ton!
添加回复
回复:Your style is really unique in comparison to other people I have read stuff from. Many thanks for posting when you've got the opportunity, Guess I'll just bookmark this site.
添加回复
回复:I read this piece of writing fully regarding the difference of most up-to-date and preceding technologies, it's amazing article.
添加回复
回复:Hey I know this is off topic but I was wondering if you knew of any widgets I could add to my blog that automatically tweet my newest twitter updates. I've been looking for a plug-in like this for quite some time and was hoping maybe you would have some experience with something like this. Please let me know if you run into anything. I truly enjoy reading your blog and I look forward to your new updates.
添加回复
回复:Good day I am so thrilled I found your blog, I really found you by mistake, while I was searching on Bing for something else, Anyways I am here now and would just like to say thanks a lot for a fantastic post and a all round interesting blog (I also love the theme/design), I don’t have time to go through it all at the minute but I have bookmarked it and also added in your RSS feeds, so when I have time I will be back to read more, Please do keep up the awesome work.
添加回复
回复:You could certainly see your skills in the work you write. The world hopes for more passionate writers like you who aren't afraid to say how they believe. All the time follow your heart.
添加回复
回复:Hi there! Someone in my Myspace group shared this website with us so I came to take a look. I'm definitely loving the information. I'm bookmarking and will be tweeting this to my followers! Exceptional blog and great design.
添加回复
回复:My brother suggested I might like this blog. He was totally right. This post truly made my day. You can not imagine just how much time I had spent for this information! Thanks!
添加回复
谐云
2025年04月19日
添加回复
谐云 CpjJwWHV
2025年04月19日
555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
回复CpjJwWHV:
添加回复
回复CpjJwWHV:
添加回复
谐云 CpjJwWHV
2025年04月19日
555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
gBqsPxAZ回复CpjJwWHV:555
添加回复
回复CpjJwWHV:
添加回复
回复CpjJwWHV:
添加回复
回复CpjJwWHV:
添加回复
回复CpjJwWHV:
添加回复
回复CpjJwWHV:
添加回复
申请合作咨询
您可以通过此表单填写您的合作意向,我们将会尽快与您取得联系!
或拨打电话0571-87607309
*姓名:
*手机:
*邮箱:
备注:
备注:
登录
登录
注册账号 忘记密码
注册
{{ code.btn }}
注册
立即登录 忘记密码?
忘记密码
{{ code.btn }}
确定
立即登录 忘记密码?
立即咨询