世界上最小的超级计算机_这些是世界上最强大的超级计算机

本文介绍全球最强大的超级计算机,如日本的Fugaku、美国的IBM Summit和中国的Sunway TaihuLight,这些计算机在科学研究、气象预测、药物设计等领域发挥着关键作用。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

世界上最小的超级计算机

You may never have seen one but supercomputers are very necessary on a day-to-day basis. You may have heard of the expression “supercomputer” but you are not sure what it refers to. These are large computers with high levels of power and performance that are used in genetic, biological, and weapons research. The power has multiplied by more than 100,000 in the last fifteen years and they are capable of working like thousands of computers together to find cures for diseases or to “guess” when a tsunami is going to hit.

您可能从未见过,但是超级计算机在日常工作中是非常必要的。 您可能听说过“超级计算机”一词,但不确定它指的是什么。 这些是具有强大功能和性能的大型计算机,用于基因,生物和武器研究。 在过去的15年中,其功能已增长了十万多倍,它们能够像数千台计算机一样工作,共同寻找疾病的根治方法或在海啸来袭时“猜测”。

但是……什么是超级计算机? (But…What Are Supercomputers?)

Image for post
IBM Summit. Source: Summit.
IBM峰会。 资料来源:峰会。

Supercomputers are computers with characteristics far superior to those we have at home that operates at a high speed, usually measured in petaFLOPS, which translates as a thousand billion operations per second. They can process billions of data per second which makes them useful for practically all kinds of tasks: from predicting a tsunami to finding a cure for coronavirus, searching for planets, etc.

超级计算机的性能远远优于我们在家中可以高速运行的计算机,通常以petaFLOPS来衡量,换算为每秒1万亿次操作。 它们每秒可以处理数十亿个数据,这对几乎所有任务都非常有用:从预测海啸到找到治疗冠状病毒的方法,寻找行星等。

In general, although we will see some of its uses later, they are used for special research, be it laboratories, education, industrial design, or drug research. They are all over the world, cost hundreds of millions of euros and it is expected that in Europe there will be many more in the coming years. Even if you are not used to seeing them, their use is essential for investigations of all kinds.

总的来说,尽管我们稍后会看到它的一些用途,但是它们被用于特殊研究,无论是实验室,教育,工业设计还是药物研究。 它们遍布世界各地,耗资数亿欧元,并且预计在未来几年中,欧洲将会有更多。 即使您不习惯看到它们,使用它们对于各种调查也是必不可少的。

它们是用来干什么的? (What Are They Used For?)

One of its most common applications is meteorology in order to face all kinds of external weather events and “anticipate” what is coming and predict weather conditions. They are capable of predicting the future climate or, for example, anticipating the effects of a tsunami to avoid a catastrophe in a country or involving human losses.

气象学是最常见的应用之一,它可以应对各种外部天气事件并“预测”即将发生的事情并预测天气状况。 他们有能力预测未来的气候,或者例如预测海啸的影响,避免在一个国家发生灾难或造成人员伤亡。

In terms of physics, they can simulate stellar evolutions or also perform other tasks such as the reproduction of aerodynamic models of what would be military aircraft. They can also be useful in other fields such as biology, medicine, physics, astronomy, geology, zoology.

在物理学方面,他们可以模拟恒星的演化,也可以执行其他任务,例如复制军事飞机的空气动力学模型。 它们在生物学,医学,物理学,天文学,地质学,动物学等其他领域也可能有用。

最常见的用途。 (Most common uses.)

Image for post
Photo by Michael Dziedzic on Unsplash
Michael DziedzicUnsplash拍摄的照片

Extending the life of nuclear weapons.

延长核武器的寿命。

Astrophysics studies.

天体物理学研究。

Sustainability, health, and climate change.

可持续性,健康与气候变化。

Meteorology.

气象。

Atmospheric composition and air quality.

大气成分和空气质量。

Big Data, the interaction of people with computers.

大数据,人与计算机的互动。

Biomechanics for simulations of respiratory, cardiovascular systems.

用于模拟呼吸,心血管系统的生物力学。

Education.

教育。

Engineering simulation.

工程仿真。

Drug design.

药物设计。

Geophysics, earthquake simulation.

地球物理,地震模拟。

Analysis of the properties of the surface of our planet.

分析我们行星表面的特性。

Bioinformatics and cynical data volume transmission.

生物信息学和愤世嫉俗的数据量传输。

Cloud computing and data center optimization.

云计算和数据中心优化。

Social simulation and study of cultural evolution.

社会模拟和文化进化研究。

他们在哪里? (Where Are They Located?)

Image for post
Photo by Shai Pal on Unsplash
Shai PalUnsplash拍摄的照片

There are more than a thousand supercomputers around the world in a large number of countries. Among them, Spain with its MareNostrum in Barcelona. But most are divided between China, the United States, and Japan, although France, Germany or the Netherlands also have some of the most important.According to the most recent data, 45.2% of the supercomputers of the 500 best supercomputers in the world. China has 226 devices out of the 500 most powerful devices on the planet.

在许多国家/地区,全球有超过一千台超级计算机。 其中,西班牙及其在巴塞罗那的MareNostrum。 尽管法国,德国或荷兰也有一些最重要的国家,但大多数国家还是分布在中国,美国和日本之间。根据最新数据,在全球500强最佳超级计算机中,超级计算机占45.2%。 在全球500个最强大的设备中,中国拥有226个设备。

联想的Marenostrum。 (Marenostrum by Lenovo.)

Image for post
Marenostrum Barcelona. Source: BSC.
巴塞罗那Marenostrum。 资料来源:BSC。

The MareNostrum computer in Barcelona occupies position 37 of the 500 most important and powerful in the world. The first model of this supercomputer was launched in 2004 when IBM (in collaboration with the Lenovo company) signed an agreement with the Government to develop this computer that initially had a power of 42.4 teraflops. But that was more than fifteen years ago.

巴塞罗那的MareNostrum计算机在全球500个最重要和功能最强大的计算机中排名37。 这款超级计算机的首款型号于2004年推出,当时IBM(与联想公司合作)与政府签署了一项协议,以开发这种最初具有42.4 teraflops功率的计算机。 但这已经超过十五年前了。

Currently, work is underway on MareNostrum 5, which is expected to be released at the end of 2020 and will have a price that exceeds 100 million euros. This computer will be 10,000 times higher in power than the 2004 model and up to 17 times higher in power than the current MareNostrum 4. It will reach 300 teraflops and is part of the European Commission’s plans to create a “network of supercomputers” in different territories of the European Union.

目前,MareNostrum 5的工作正在进行中,预计将于2020年底发布,价格将超过1亿欧元。 该计算机的功率将比2004年的型号高10,000倍,比当前的MareNostrum 4的功率高17倍。它将达到300 teraflops,并且是欧洲委员会计划在不同领域创建“超级计算机网络”的计划的一部分。欧洲联盟领土。

这些是前五名。 (These Are The Top Five.)

The most powerful computers in the world are continuously updated in the Top500 list compiled by researchers around the world since 1993. It is updated twice a year in case a new supercomputer has ousted another from the list.

自1993年以来,世界上功能最强大的计算机一直在由世界各地的研究人员编制的Top500列表中不断更新。万一新的超级计算机从该列表中淘汰另一台超级计算机,则每年更新两次。

富士通的超级计算机Fugaku。 (Supercomputer Fugaku by Fujitsu.)

Image for post
Supercomputer Fugaku. Source: DataCenterDynamics.
超级计算机Fugaku。 资料来源:DataCenterDynamics。

Japan has become number one on the list of the most powerful computers in the world thanks to Fugaku, a computer tested in June 2020 that has broken all the statistics measured to date. The Japanese supercomputer has reached a Rmax of 415,530 TFlops / s and a Rpeak of 513,584.7 TFlops / s. The beastly Japanese machine has been ranked as the most powerful in the world thanks to these figures and is located at the RIKER Center for Computational Science research institute, in Kobe.

得益于Fugaku,日本已成为世界上功能最强大的计算机列表中的第一名。Fugaku于2020年6月进行了测试,打破了迄今为止测得的所有统计数据。 日本超级计算机的Rmax达到415,530 TFlops / s,Rpeak达到513,584.7 TFlops / s。 由于这些数据,这台令人震惊的日本机器被评为世界上最强大的机器,位于神户的RIKER计算科学研究院中心。

IBM峰会。 (Summit by IBM.)

Image for post
Summit. Source: IBM.
首脑。 资料来源:IBM。

Before the appearance of Fugaku, this American supercomputer held the world leader in its hand. It is located at Oak Ridge National Laboratory in Tennessee and is powered by IBM POWER9 and NVIDIA Tesla V100 processors. Its Rmax is 148,600 TFlops / s and its Rpeak is 200,794.9 TFlops / s. Summit has been dedicated to identifying patterns that could anticipate the diagnosis of Alzheimer’s disease, as well as other genetic or meteorological investigations.

在Fugaku出现之前,这台美国超级计算机牵手了世界领先者。 它位于田纳西州的橡树岭国家实验室,由IBM POWER9和NVIDIA Tesla V100处理器提供支持。 其Rmax为148,600 TFlops / s,其Rpeak为200,794.9 TFlops / s。 Summit致力于确定可预测阿尔茨海默氏病诊断的模式,以及其他遗传或气象研究。

Sierra,同样由IBM提供。 (Sierra, Also by IBM.)

Image for post
Sierra. Source: IBM.
塞拉 资料来源:IBM。

Number three on the list of the most powerful supercomputers in the world is for this machine from IBM, Nvidia, and Mellanox that has achieved a Rmax of 94,640.0 TFlops per second and a Rpeak of 125,712 TFlops per second. It is located at the Lawrence Livermore National Laboratory in California and has conducted research to ensure the safety of US nuclear weapons, as well as their reliability.

对于这台来自IBM,Nvidia和Mellanox的机器而言,世界上功能最强大的超级计算机中排名第三的Rmax为94,640.0 TFlops /秒,Rpeak为125,712 TFlops /秒。 它位于加利福尼亚州的劳伦斯·利弗莫尔国家实验室,并进行了研究以确保美国核武器的安全性及其可靠性。

NRCPC的Sunway TaihuLight。 (Sunway TaihuLight by NRCPC.)

Image for post
Sunway. Source: NRCPC.
双威 资料来源:NRCPC。

China is left out of the current podium and has to settle for fourth place in this TOP 500 that classifies the most powerful supercomputers in the world. Despite owning most of the machines that make up this TOP 500, the Chinese are left without the first three seats and their first machine is this Sunway TaihuLight which is located in the Wuxi national supercomputing center. Its data says it returns a Rmax of 93,014.6 TFlops per second and a Rpeak of 125,435.9 TFlops per second.

中国被排除在目前的领奖台之外,必须跻身全球500强超级计算机前500名之列。 尽管拥有构成TOP 500的大多数机器,但中国人却没有前三个席位,而他们的第一台机器是位于无锡国家超级计算中心的Sunway TaihuLight。 它的数据表明,它每秒返回的Rmax为93,014.6 TFlops,每秒的峰值为125,435.9 TFlops。

NUDT开发的Tianhe-2A。 (Tianhe-2A by NUDT.)

Image for post
Tianhe 2A. Source: NUDT.
天河2A。 资料来源:NUDT。

Another Chinese computer that is among the most powerful computers in the world is this Tianhe-2A that has Intel Xeon E5–2692 v2 processors and whose power data is 61,444.5 and 100,678.7 TFlops / s in Rmax and Rpeak respectively. It made it to number 1 on the list and was the fastest computer in the world between June 2013 and June 2016. It is located in Guangzhou.

天河2A是世界上功能最强大的计算机之一,它具有Intel Xeon E5–2692 v2处理器,其功率数据的Rmax和Rpeak分别为61,444.5和100,678.7 TFlops / s。 它在榜单上排名第一,并且是2013年6月至2016年6月之间世界上最快的计算机。它位于广州。

Read more Medium Stories.

阅读更多中等故事。

翻译自: https://medium.com/age-of-awareness/these-are-the-most-powerful-supercomputers-of-the-world-69d126f1a4da

世界上最小的超级计算机

### 超级计算机计算资源的限制 超级计算机作为当前强大的计算设备之一,在其设计和运行过程中仍然受到多种因素的影响,这些影响主要体现在硬件架构、能耗管理、散热效率以及算法优化等多个方面。 #### 硬件架构的局限性 尽管超级计算机拥有极高的并行处理能力和大规模存储支持,但由于基本计算单元之间的通信延迟较高,可能导致数据传输瓶颈。这种现象会降低整体系统的实际性能表现[^1]。此外,随着节点数量增加,互连网络复杂度也会相应提升,从而进一步加剧这一问题。 #### 能耗与散热挑战 高性能运算需要消耗大量电力资源,而过高的功耗不仅增加了运营成本,还带来了严重的热管理难题。如果无法有效解决温度控制问题,则可能损害硬件组件甚至引发系统崩溃风险。因此,在构建新一代超算平台时必须充分考虑能效比优化措施[^2]。 #### 浮点运算能力及其评估标准 对于涉及科学仿真或者工程分析等领域内的任务来说,浮点操作往往占据主导地位。此时我们常用FLOPS来描述目标机器针对此类工作负载所能达到的大吞吐量水平[^4]。然而值得注意的是,理论峰值并不总是能够完全转化为实际应用中的收益;因为除了单纯考察单核效能之外,还需要关注整个集群内部协作机制是否高效合理等问题[^3]。 ```python # 示例代码展示如何简单估算某款处理器潜在浮点性能(假设已知频率GHz及每次时钟可完成多少次双精度FP ops) def estimate_flops(clock_speed_ghz, flops_per_cycle): return clock_speed_ghz * 1e9 * flops_per_cycle example_clock = 3.0 # GHz flops_each_tick = 8 estimated_performance = estimate_flops(example_clock , flops_each_tick ) print(f"Estimated peak performance: {estimated_performance} FLOPs") ```
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值