英伟达驱动的超级计算机推动抗新冠药物的发现和原子模拟

Supercomputers built using Nvidia Corp.’s most advanced platforms are leading the fight against the coronavirus pandemic, helping researchers gain more insights into the nature of the SARS-CoV-2 virus and creating new artificial intelligence models to accelerate drug discovery.

英伟达（Nvidia）最先进的平台可用于打造超级计算机，这些超级计算机现在对于对抗冠状病毒大流行起到重要的领导作用，它们可以帮助研究人员更深入地了解2019新冠病毒（COVID-19）的特性及创建新的人工智能模型和加速发现抗冠状病毒药物。

Nvidia revealed today how a team led by Arvind Ramanathan, a computational biologist at the Argonne National Laboratory, designed a new workflow to study the virus that runs across multiple supercomputing systems, including the Nvidia A100 graphics processing unit-powered Perlmutter. Ramanathan’s researchers have created a way to improve the resolution of traditional tools used to explore DNA, creating fresh insights that may help to arrest the spread of COVID-19.

根据Nvidia今天发布的消息，Argonne国家实验室的计算生物学家Arvind Ramanathan领导的团队正在设计一个新的病毒研究工作流程，该流程在多个超级计算系统中运行，包括Nvidia A100图形处理单元驱动的超级计算系统Perlmutter。Ramanathan手下的研究人员开发的方法可以提高用于探索DNA的传统工具的分辨率并可以得到可能有助于阻止COVID-19传播的新见解。

“The capability to perform multisite data analysis and simulations for integrative biology will be invaluable for making use of large experimental data that are difficult to transfer,” Ramanathan’s researchers said.

Argonne国家实验室的研究人员表示，“能够利用多站点数据分析和模拟这种综合生物功能对于充分利用那些难于传送的大型实验数据将是非常宝贵的。”

The work included the development of a new technique that makes it possible to accelerate molecular dynamics research by running the Nanoscale Molecular Dynamics program on GPUs. Combined with Nvidia’s NVLink networking technology, the team was able to process data far beyond what was previously possible with conventional high-performance computing interconnects.

Argonne国家实验室的这项工作包括开发一项新技术，使其可以通过在GPU上运行纳米级分子动力学程序加速分子动力学的研究。实验室团队结合Nvidia的NVLink网络技术处理的数据远超过去传统的高性能计算互连可能做得到的。

In a separate project, a team led by Ivan Oleynick from the University of Florida created what they say is the world’s first simulation of a billion atoms using Nvidia-powered machines. The teams says the simulation of carbon atoms under extreme temperature and pressures could open doors to new energy sources and help astronomers to discover the makeup of distant exoplanets. They added that the simulation offers quantum-level accuracy that faithfully reflects the forces among each of the billion atoms.

另外一个项目，佛罗里达大学的Ivan Oleynick带领的团队使用Nvidia驱动的机器打造出他们说是世界上第一个10亿原子的模拟系统。两个团队都表示，在极端温度和压力下成功模拟碳原子可以打开新能源的大门，也可以帮助天文学家发现遥远的系外行星的构成。他们补充表示，该模拟系统的准确性达到了量子级别，模拟系统能忠实反映出十亿个原子之间的力场。

“It’s accuracy we could only achieve by applying AI techniques,” Oleynik said.

Oleynik表示，“这种精确度我们只能通过用人工智能技术才能达到。”

The simulation was created using an InfiniBand-connected system made up of a stunning 27,900 GPUs on the Summit supercomputer built by IBM Corp. The team demonstrated that it’s possible to scale the simulation to 20 billion or more atoms if required.

该模拟系统是基于InfiniBand连接系统创建的，InfiniBand连接系统由洋洋27900个GPU组成，置于由IBM公司打造的Summit超级计算机里。该团队的研究表明，如果有需要的话，模拟规模有可能扩展到200亿或更多的原子。

A second billion-atom simulation, meanwhile, was carried out by a team led by researcher Rommie Amaro at the University of California at San Diego. That effort, which also relied on Summit’s computing power, was aimed at improving understanding of COVID-19 by simulating the Delta variant in an airborne droplet. The work is said to have provided insights into how the virus binds itself in the deep lung, and could also aid our understanding of the progression of severe diseases such as cancer and cystic fibrosis, the team said.

与此同时，加州大学圣地亚哥分校研究员Rommie Amaro的带领团队在进行第二个10亿个原子的模拟。这项工作也依赖Summit的算力，项目的目的是模拟空气中雾滴里德尔塔变异株并提高对COVID-19的理解。该团队称，这项工作可望提供关于此类病毒如何在肺部深处粘合自身的一些见解，并且还可以帮助我们了解癌症和囊性纤维化等严重疾病的恶化过程。

“We demonstrated how AI coupled to HPC at multiple levels can result in significantly improved effective performance,” the paper said.

团队发布的论文表示，“我们得以展示人工智能如何在多个层面上与高性能计算机结合能够显著提高有效性能。”

img_

The Amaro team simulated a coronvirus in an airborne droplet with a billion atoms

Amaro团队用10亿个原子模拟空气中雾粒子里的冠状病毒

One final effort involved applying natural language processing to the problem of screening chemical compounds for new drugs. Jens Glaser, a computational scientist at Oak Ridge National Laboratory, said his team used 24,000 Nvidia GPUs on Summit to train a BERT NLP model that’s able to speed up drug discovery on a dataset containing 9.6 billion molecules, in just two hours. Previously, it took up to four days to train similarly sized models, Nvidia said.

本文介绍的最后一项工作涉及到将自然语言处理应用到筛选新药化合物。Jens Glaser是橡树岭国家实验室的计算科学家。他表示，他的团队在Summit上用24,000个Nvidia GPU训练一个BERT NLP模型，该模型能够在含96亿个分子的数据集上加速药物发现，用时仅需两小时。Nvidia公司称以前训练类似规模的模型需时四天。

“We’re just scratching the surface of training data sizes — we hope to use a trillion molecules soon,” said Andrew Blanchard, a research scientist who led the team.

带领该团队的研究科学家Andrew Blanchard表示，“我们在训练数据规模上只是小试牛刀，我们希望很快能用上一万亿个分子。”

All four projects have been nominated for a prize at the Gordon Bell awards, which are widely considered to be the equivalent of a Nobel prize in high-performance computing.

所有四个项目都被提名参加Gordon Bell奖的评选，业内人士广泛认为该奖项相当于高性能计算领域的诺贝尔奖。

### Nvidia’s newest InfiniBand accelerates HPC

Nvidia最新的InfiniBand加速了高性能计算的发展

In other news revealed today, Nvidia said the academic research community has wasted little time in putting its latest supercomputer networking technology to use. Two universities have announced plans to plug into the Nvidia’s new Quantum-2 InfiniBand platform that was unveiled during last week’s GTC 2021 event.

Nvidia在今天发布的其他新闻里表示，学术研究界很快就将用上Nvidia旗下最新的超级计算机网络技术。两所大学已经宣布切入Nvidia的新Quantum-2 InfiniBand平台的计划，新的Quantum-2 InfiniBand平台是在上周的GTC 2021活动上公布的。

Nvidia’s InfiniBand is a computer networking communications standard used in high-performance computing for data flow between processors and input/output devices, known for its high throughput and very low latency.

Nvidia的InfiniBand是一种计算机网络通信标准，主要用于高性能计算中处理器和输入输出设备之间的数据流，该标准以高吞吐量和极低的延迟著称。

img_

The latest generation of InfiniBand, Nvidia Quantum-2, introduces new features that will accelerate the most demanding workloads in supercomputers. Quantum-2 doubles network speed while tripling the number of network ports available, meaning overall performance can be accelerated by up to three-times compared with existing systems.

最新一代的InfiniBand是Nvidia Quantum-2，Nvidia Quantum-2引入了新功能，可以加速超级计算机中最苛刻的工作负载。Quantum-2的网络速度提高了一倍，同时Quantum-2的可用网络端口数量提高了三倍，这意味着与现有系统相比，Quantum-2整体性能可加快三倍。

Among the first to deploy Quantum-2 InfiniBand will be Texas A&M. The university said it’s planning to use the 400G InfiniBand network in its ACES supercomputer to connect researchers to a mix of five accelerators from four vendors.

德州农工大学是首批部署Quantum-2 InfiniBand的大学。德州农工大学表示正计划在大学的ACES超级计算机中使用400G InfiniBand网络，ACES超级计算机可将研究人员与来自四家供应商的五个加速器混合在一起。

“Besides the obvious two-times jump in throughput from Nvidia Quantum-1 InfiniBand at 200G, it will provide improved total cost of ownership, beefed up in-network computing features and increased scaling,” said Honggao Liu, ACES’s principal investigator and project director.

ACES的主要研究人员和项目主任Honggao Liu表示，“Quantum-2 InfiniBand除了在吞吐量上比200G的Nvidia Quantum-1 InfiniBand明显跃升两倍之外还将提供更优的总体拥有成本及加强网络内的计算功能和提高扩展性。”

The other early adopter is Mississippi State University, which is planning to deploy Quantum-2 InfiniBand to expand its Orion supercomputer that runs massive weather forecasting jobs for the U.S. National Oceanic and Atmospheric Administration. The system was ranked as the fourth-largest academic supercomputer in the U.S. when it first went live in June 2019.

Quantum-2 InfiniBand的另一个早期采用者是密西西比州立大学，密西西比州立大学正计划部署Quantum-2 InfiniBand扩展学校的Orion超级计算机，Orion超级计算机为美国国家海洋和大气管理局运行大量的天气预报工作负载。该系统在2019年6月首次上线时被列为美国第四大学术超级计算机。

“HPC is going everywhere,” said Gilad Shainer, senior vice president of marketing at Nvidia. That, he noted, will require the kind of cloud-native supercomputing Quantum-2 brings. “It’s bringing supercomputing into the cloud,” he said.

Nvidia营销高级副总裁Gilad Shainer表示，“高性能计算将比比皆是。”他指出，比比皆是的高性能计算需要随Quantum-2而来的那种云原生超级计算，“这样就可以将超级计算带入云端”。

### Europe gets a new AI research lab

欧洲新建人工智能研究实验室

Nvidia also revealed that it has partnered with the French information technology giant Atos SE on an initiative aimed at advancing European computing technologies, education and research. The new Excellence AI Lab, or EXAIL, is building an exascale-class supercomputer that will be powered by Nvidia’s GPUs and Arm-based Grace central processing units, with Quantum-2 InfiniBand networking and Atos’ BXI Exascale Interconnects.

Nvidia还透露正在与法国信息技术巨头Atos SE合作开展一项旨在推进欧洲计算技术、教育和研究的活动。新的卓越人工智能实验室EXAIL正在打造一台超级规模计算机，该超级规模计算机将由Nvidia的GPU和基于Arm的Grace中央处理单元驱动，采用Quantum-2 InfiniBand网络和Atos的BXI Exascale互连。

The lab is planning to accelerate research into five key areas where it believes its new supercomputer will be able to make a real impact: climate research, healthcare and genomics, hybridization with quantum computing, edge AI and computer vision, and cybersecurity.

EXAIL实验室正计划加快在五个关键领域的研究，EXAIL认为自己的新超级计算机能够在下列五个关键领域产生真正的影响：气候研究、医疗保健和基因组学、量子计算混合、边缘人工智能和计算机视觉以及网络安全。