Redian新闻
>
云计算:芯片大战必争之地​ ​| 经济学人商业

云计算:芯片大战必争之地​ ​| 经济学人商业

公众号新闻

1



写在前面

01 免费外刊分享群 
最近很多卖外刊的平台/网盘都挂了,后台有小伙伴私信小编问我们这里还免费分享外刊吗?小编的回答是:必须的!虽然已经发过多次推文,例如获取经济学人的N种方法如何在官网半价订阅经济学人,讲过了这么多方法,但还是有辣么多小伙伴不会或者懒得弄。于是小编再次建了群,会每天以及每周在群里分享外刊,有兴趣的小伙伴扫码进群(已经加入公众号其他群的小伙伴请不要重复加群),谢谢。如果群满了,请转发这篇文章到朋友圈,私信小编微信foxwulihua2即可加入进群。进群发广告者一律踢,不接受任何解释,谢谢理解。扫以下二维码进群(已更新):


1. 每个工作日早上群里分享The world in brief 
为什么要读the world in brief?引用翻译组裸考过一笔经验贴 中写的一段话:每天精读经济学人EspressoEspresso相当于经济学人的微缩版,每周六天,每天五篇,给阅读经济学人提供了必要的背景知识。每天看有连贯性,如同看一部连续剧,而世界就是这个大舞台。

2.每周末会随机分享多种外刊PDF 


02 第十三期翻译打卡营 

4位一笔,3位二笔
21天录播课程
既有中译英,也有英译中。
从翻译的流程策略,英汉语言特点,
方法,类型,背景知识到细致的语言点,
我们都逐一讲解以及答疑,让同学们吃透翻译。
点击下图,即可了解课程详情!


2



听力|精读|翻译|词组
Business | Schumpeter
商业 熊彼特专栏
英文部分选自经济学人20221008期商业版块

Business | Schumpeter
商业 | 熊彼特专栏

The cloud is the fiercest front in the chip wars
云计算:芯片大战必争之地

Data-centre chips were once a stale monopoly. Now the business is brimming with competition
数据中心芯片曾垄断如一潭死水 如今竞争激烈
 
It is easy to think of the computing cloud as the placeless whereabouts of the latest Netflix series, your Spotify playlists, millions of wanton selfies and your digital assistant. It is even easier to ignore it altogether, at least until Alexa alerts you that your storage space is filling up and helpfully offers to rent you extra room, of which there always appears to be more available. Necessary, disembodied and, for $9.99 a month, to all intents and purposes limitless: it is the ether of the digital age. This ether, though, has a very unethereal side—the vast data centres where all this information is physically stored and, increasingly, processed by powerful computers known as servers. The semiconductor hardware that makes the servers powerful is fast becoming the hardest-fought front in the battle over the $600bn global market for computer chips.

云端往往被认为是用于存放Netflix最新剧集、Spotify播放列表、数百万张恣意自拍照的虚拟空间,你的个人智能助手也“住”在那儿。而更多时候,“云端”根本不会被人想起,除非Alexa(亚马逊开发的智能语音助手)提醒你存储空间已满,邀请你升级存储空间,并且这些存储空间好像总是取之不尽、用之不竭。每月支付9.99美元,便能获得必要且近乎无限的虚拟存储空间,这就是数字时代的“以太”。然而,这一“以太”却有现实的真身,即庞大的实体数据中心,用于储存所有信息,并且越来越多强大的计算机(即“服务器”)被用来处理这些信息。在6000亿美元规模的全球计算机芯片市场中,服务器强大计算能力背后的半导体硬件正迅速成为争夺最激烈的战场。

注释:
1placeless: not in, or connected with, any particular place. 没有固定位置的
2whereabouts: the place where a person or thing is. 下落,所在之处
3ether: 以太,古希腊哲学家亚里士多德所设想的一种物质,在亚里士多德看来,物质元素除了水、火、气、土之外,还有一种居于天空上层的以太。 


Rooms of servers began to replace computer mainframes in the 1990s. Back then, they were owned by companies and installed on their premises. They mostly ran on chips made byIBM and HP, the big tech of the day. These were supplanted by processors from Intel, which by the mid-2000s translated its dominance of PC semiconductors into a near monopoly of the server market. Things started to change once again around a decade ago, when Amazon began selling some of its spare server capacity. Microsoft and Google followed suit and the cloud-computing industry took shape. As the cloud has billowed, so has Intel’s competition. 

20世纪90年代,服务器机房开始取代计算机大型机。当时,各大公司在各自的场地里有自己的大型机,其运行大多基于当时的科技巨头IBM惠普制造的芯片,后来转而使用英特尔的处理器。到2000年代中期,英特尔从统治个人电脑半导体市场,发展至近乎垄断服务器市场。大约10年前,情况再次发生变化:亚马逊开始出售部分闲置服务器容量,微软和谷歌纷纷效法,云计算行业初见雏形,英特尔面临的竞争也随“云”的升腾而潮起云涌。

Today the market for server processors is getting bigger, more crowded and more complex. (The business of memory chips, which store data rather than crunch it, is distinct, more commoditised and less lucrative.) Intel, which both designs and manufactures semiconductors, derives 33% of its revenues from server chips, up from 29% in 2016. Specialist chip designers that do not do their own manufacturing are expanding their server-chip businesses even more quickly. Data centres now account for 39% of the sales of Nvidia, up from 7% six years ago. ForAMD, another American chip designer, the figure has jumped from 17% to 23% between 2020 and 2021. The big cloud providers, including Amazon and Google, are getting in on the game by designing their own processors, often based on blueprints from Arm, a Japanese-owned firm which licenses off-the-shelf designs that clients can tailor to their needs. The designs are then etched onto silicon by contract manufacturers, most notably TSMC of Taiwan, which are also increasing their server-chip capacity.

如今,服务器处理器市场规模越来越大,竞争对手越来越多,竞争形势愈加复杂。而存储芯片业务与之截然不同,因为存储芯片仅存储数据,而非快速处理数据,因此商品化程度更高,利润也更低。英特尔公司既是半导体设计商又是制造商,其33%的收入源于服务器芯片,高于2016年的29%。相比之下,专业芯片设计商(只设计不生产)的服务器芯片业务发展势头更猛。如今,数据中心业务占英伟达销售总额的39%,高于6年前的7%。另一家美国芯片设计商超威半导体公司(AMD)的该项业务销售额在20202021期间从17%跃升23%。亚马逊和谷歌等大型云供应商也纷纷自主研发处理器,加入市场竞争行列,他们通常使用日本安谋Arm)公司批准的现成设计蓝图来定制所需产品。代工厂根据设计用硅制造出芯片,最出名的代工厂的要数台积电(TSMC),该公司目前也在持续扩大服务器芯片产能。

注释:
Arm: 软银旗下芯片设计公司。Arm公司是专门从事基于RISC技术芯片设计开发的公司,作为知识产权供应商,本身不直接从事芯片生产,而是转让设计许可,由合作公司生产各具特色的芯片

Two factors explain the competitive storm. The first is the market’s size and growth. Data-centre chips are a bright spot in an otherwise dark year for the semiconductor industry, which is in the grip of a cyclical downturn that has wiped two-fifths from the market value of the world’s chipmakers this year. Whereas sales of pcs and smartphones, and the chips inside them, are expected to fall this year, server demand is forecast to rise. Synergy Research Group, a firm of analysts, expects the cloud giants to build more than 300 new data centres around the world by 2024. The biggest of these could house at least 100,000 servers apiece. That will require an awful lot of chips. According to IDC, another research firm, cloud and on-premise data centres will buy $71bn-worth of semiconductors in 2022, up from $42bn in 2019. Sales may grow by nearly half in the next five years or so, twice as fast as the chip industry as a whole. Most of that growth will come from processors rather than memory chips.

这场竞争风暴的形成有两方面原因。首先在于市场的规模和增长。半导体行业正在经历周期性低迷,今年全球芯片制造商的市值也因此蒸发了五分之二,而数据中心芯片正是该行业至暗一年里的一道曙光。尽管今年个人电脑和智能手机及其芯片销量预计将下降,但服务器需求预计还会上升。分析机构Synergy Research Group预计,到2024,云计算巨头将在全球新建300多家数据中心,其中规模最大的可容纳至少10万台服务器,而这产生了巨大的芯片需求。另一家研究公司IDC的数据显示,2022年,云数据中心和内部数据中心将产生价值710亿美元的半导体需求,高于2019年的420亿美元。未来5年左右,数据中心芯片销售额将增长近50%,是整个芯片行业销售额增速的两倍。其中大部分增长源于处理器芯片,而非存储芯片。

The second reason for the upheaval is the growing sophistication of what the cloud does. It no longer acts merely like a large external hard drive. It is bursting with new capabilities that require different chip architectures. In some cases, that means repurposing existing technology. Nvidia’s cloud business is built atop its graphics processing units (GPUS), specialised chips used to make computer animation lifelike. It turns out thatGPUS, which were first designed in the 1990s to improve video games, are also excellent at running artificial-intelligence (AI) models. Intel recently launched its first set of stand-alone GPUS to compete with Nvidia as well as AMD, which also makes them. 

此外,这场竞争风暴还归咎于日渐复杂的云服务功能。云端不再仅是大容量外部硬盘驱动器,它也具备了多种新功能,需要不同的芯片架构。这有时意味着需要重构现有技术。英伟达的云业务基于其图形处理器(GPUS),这种专用芯片可以让电脑动画栩栩如生。事实表明,这些20世纪90年代首次推出,最初旨在优化电子游戏的图形处理器,如今在运行人工智能模型方面也表现优异。近期,英特尔推出首批独立图形处理器,意与同样生产GPU的英伟达和超微半导体公司抗衡。

Cumulo-nimble
云服务之战

All-new designs are also emerging. The cloud giants, looking to boost performance and cut costs, are busily adapting Arm’s energy-efficient designs. Amazon’s Arm-derived Graviton chips are being sown across many of its server farms. Google is doing the same with its Tensor Processing Units. Microsoft is working on custom designs for its Azure cloud. In 2020 Nvidia offered to buy Arm for $40bn, in large part to beef up its cloud offering. That deal collapsed in February amid antitrust scrutiny, but the company nevertheless plans to launch a general-purpose server chip next year to take on Intel more directly. 

与此同时,全新的设计方案也在不断涌现。云服务巨头公司既想提高性能,又想压低成本,纷纷忙于改造Arm的低能耗设计。亚马逊在许多服务器群组里嵌入了衍生自ArmGraviton芯片;谷歌也在自己的服务器群组内用上了自研的张量处理单元(Tensor Processing Units);微软正致力于为Azure云平台量身定制芯片设计。2020年,英伟达欲以400亿美元收购Arm公司,主要旨在加强其云服务功能。但今年2月,该收购计划因反垄断审查破产。即便如此,英伟达依旧计划明年推出一款通用服务器芯片,用以与英特尔直接竞争。

注释:
Tensor Processing Units,张量处理单元,简称TPU,是一种专为机器学习开发的专用集成电路。相较于传统 GPU,其设计很明确,主要是通过降低计算精度以提高效能,且没有硬件光栅和纹理映射单元。TPU 芯片是专为谷歌的 TensorFlow 设计的,但目前谷歌仍将 CPU GPU 用于其他类型的机器学习,目前除 TPU 以外,还有主要用于嵌入式和机器人领域的 AI 极速器 AI Accelerator

One clear winner from the cloud boom is TSMC. It is the only company currently able to manufacture the bleeding-edge processors most sought-after by the cloud providers. Intel hopes that its recent technological advances and entry into the contract-manufacturing business—together with $52bn in new American subsidies for domestic chipmaking—will help close the gap with the Taiwanese company and win back some of that custom (though the firm’s consistently underperforming share price suggests that investors have their doubts).

台积电显然是这场云计算热潮中的胜者。它是目前唯一能够生产云服务供应商急需的尖端处理器的公司。英特尔希望凭借其最近的技术进步和新开展的代工制造业务,以及美国对国内芯片制造业新发放的520亿美元补贴,能够帮助其缩小与台积电的差距,并赢回一些客户(不过,英特尔持续萎靡的股价表明投资者仍心存顾虑)。

Ironically for a hardware business, another big chunk of the cloud-chip spoils may end up with firms that offer the best software. Nvidia’s popular programming language,CUDA, already makes it easier for developers to boost the performance of its chips. For now, the cloud giants seem content to work with Nvidia rather than try to compete with its specialised software. But they are first and foremost software firms, so this peaceful co-existence may not last for ever. All this should worry incumbents like Intel and Nvidia. For cloud users, it almost certainly means better, cheaper and ever more invisible services. 

然而讽刺的是,对硬件企业而言,这场云芯片之争的另一大战利品可能最终会落入提供最佳软件的公司手中。英伟达发布的编程语言CUDA应用广泛,已经降低了开发商提高其芯片性能的难度。目前,相较于跟英伟达的专业软件竞争,云计算领域的巨头们似乎乐意与英伟达保持合作关系。但它们毕竟是软件公司,所以和平共处的状态可能不会持续太久。因此,英特尔和英伟达等老牌企业有理由感到担忧。而对于云服务用户来说,行业竞争几乎一定会带来更优质、更廉价、更无形的服务。

翻译组:
Devin,荣枯有数,得失难量
Eva,寻路中,偶尔怀念,时常向前
SummerQE在职,梦想能仗翻译/音乐/健康走天涯
Octavia,键盘手和古风爵士,逃离舒适圈,有只英短叫八爷

校对组:
Alison贪玩又自由的风筝
Lucian,医学翻译,灵魂在杰作中冒险
Harold,工作打牌读书足球不可兼得,我偏全要

3





打造
独立思考 | 国际视野 | 英文学习
小组


01 第十三期翻译打卡营 

4位一笔,3位二笔
21天录播课程
既有中译英,也有英译中。
从翻译的流程策略,英汉语言特点,
方法,类型,背景知识到细致的语言点,
我们都逐一讲解以及答疑,让同学们吃透翻译。
点击下图,即可了解课程详情!


02 第二期考研写作课 

写作团队精研多年考研作文,

从中梳理出大小作文的写作思路,

并以评分标准为指导,

提炼出一套有效的写作能力提升方案

旨在帮助同学们高效地提高写作分数。

有思路、有句型、有词汇,文章自然水到渠成

点击下图,即可了解写作课详情!


03 第八期外刊精读课 

想要读懂更多外刊,

尽在第八期外刊精读

从字词-逻辑结构-背景-专业性答疑,

从预习-精读-泛读,全方位训练英语思维,

带你转外刊!两期连报,价格更低哦!

点击下图,即可了解精读课详情!


04 早起打卡营 

15个月以来,小编已经带着8000多人早起打卡
早起倒逼自己早睡,戒掉夜宵,戒掉手机
让你发现一个全新的自己,创造早睡早起的奇迹!
早起是最简单的自律!
第34期六点早起打卡营
欢迎你的加入!
点击下图,即可了解早起打卡营详情

微信扫码关注该文公众号作者

戳这里提交新闻线索和高质量文章给我们。
相关阅读
“超前瞻”苹果秋季发布会:芯片是一切创新驱动力宁可少活二十年拼命拿下大油田维州大选16个边缘选区公布,包括Box Hill!华人区成必争之地买房风波(3)钙离子通道(Calcium Channel)与心脏病云计算风向标:自研芯片重构云上算力云原生芯片创新加速,亚马逊云科技重构云计算格局高性能计算:RoCE技术分析及应用双11备战必看!四大品牌场域协同实操案例带来新解法|DOUCASE办公室“考古记” ​| 经济学人商业【经济学入门】顶尖文理学院教授,多项经济学论文竞赛获奖导师带你感悟经济学原理!看看排名前十的高薪工作有哪些关于芯片大大的坏消息?不是,是《中国芯片的正道何在?》三位经济学家分获2022年诺贝尔经济学奖躁动图计算:蚂蚁和字节们想找到“幻视”额头上那颗宝石佐州不愧是商家必争之地,连续第 9 年被评为全美商业第一州七宗罪与职场​ ​| 经济学人商业老板化身打工仔 ​| 经济学人商业自动增量计算:构建高性能数据分析系统的任务编排包括美联储前主席在内的三位美国经济学家分享2022年诺贝尔经济学奖!去海外搞钱,下一个必争之地在哪里?天蓝和玫红2022诺贝尔经济学奖授予美联储前主席伯南克等3位经济学家​想请病假,难上加难 ​| 经济学人商业科技投资浪潮褪去,到底谁在裸泳? ​| 经济学人商业分析师预警:芯片业凛冬将至高精度地图成智能汽车必争之地如何看待游戏化 ​| 经济学人商业经济调节的无力感--向老天爷低头 | 经济学人财经中东为何成了音乐巨头的“兵家必争之地”?《智本社经济学框架》:探索正确的经济学亚马逊试水医疗保健业 ​| 经济学人商业关于混合办公的若干问题 ​| 经济学人商业储能器件也可以计算:一种新型的神经形态器件 | NSR田丰、严飞对谈直播 | 欲望与计算:我们的消费经验史
logo
联系我们隐私协议©2024 redian.news
Redian新闻
Redian.news刊载任何文章,不代表同意其说法或描述,仅为提供更多信息,也不构成任何建议。文章信息的合法性及真实性由其作者负责,与Redian.news及其运营公司无关。欢迎投稿,如发现稿件侵权,或作者不愿在本网发表文章,请版权拥有者通知本网处理。