斯图加特Hawk超级计算机添加192块英伟达GPU 助力人工智能

斯图加特高性能计算中心（又名HLRS，德语H chstleistungsrechenzentrum Stuttgart的缩写）日前为中心的Hawk超级计算机添加了近200块英伟达图形处理器（GPU）。HLRS在此以前已经是欧洲最快的工业生产通用计算机了。

The Hawk was installed earlier this year and is seen as HLRS flagship supercomputer. Its based on a Hewlett Packard Enterprise Co.-Apollo system and currently ranks as the 16th most powerful supercomputer in the world. It's also the fastest system in Europe thats built purely on a central processing unit architecture.

Hawk是今年早些时候安装的，也是HLRS的旗舰超级计算机。Hawk基于惠普企业有限公司的阿波罗系统，目前在世界上超级计算机榜单上排第16位。Hawk亦是欧洲最快的纯粹构建在中央处理器架构上的系统。

That will soon change, though, as Hawk is about to be turbocharged with 192 of Nvidias latest A100 graphics processing units linked via a Mellanox InfiniBand network. The A100 was launched earlier this year and is said to be 20 times more powerful than Nvidias previous GPU, the Volta. Its designed for the most advanced artificial intelligence, data analytics and cloud-based graphics workloads.

不过Hawk在榜单上的排位即将发生变化，事关Hawk将添加192块英伟达最新的A100 GPU，这些GPU通过Mellanox的InfiniBand网络连在一起。A100是今年早些时候推出的，据说比以前的英伟达Volta强大20倍。A100主打最新潮的人工智能、数据分析和基于云的图形工作负载。

HLRS said the expanded Hawk supercomputer will be used for research projects that range from predicting the spread of COVID-19 to doing the science behind building better cars and airplanes.

HLRS表示，扩展后的Hawk超级计算机将用于各种研究项目，包括预测Covid-19的传播和一些涉及建造更好的汽车和飞机的科学研究项目。

Humans can create huge simulations, but we cant always understand all the data," said HLRS Director Michael Resch. "The big advantage of AI is it can work through the data and see its consequences."

HLRS中心主任Michael Resch表示， 研究人员可以创建巨大的模拟系统，但我们有时理解不了所有的数据。AI的最大优势是可以处理这些数据及看到结果。

HLRS already has one Nvidia-powered AI supercomputer, a Cray CS-Storm system installed in 2019 thats powered by 60 Nvidia Volta GPUs. Its used to run AI programs that analyze market data for Mercedes-Benz AG, a large German investment bank and other customers.

HLRS的另一个Cray CS-Storm系统也是个装了英伟达GPU的AI超级计算机，是在2019年安装的，含60块英伟达Volta GPU。这台AI超级计算机用于为梅塞德斯-奔驰公司以及一家大型德国投资银行和其他客户分析市场数据的一个AI项目。

Resch said customers were demand even greater AI processing power, hence the decision to update Hawk. The enhanced machine will be able to power hybrid computing workloads, with GPUs and CPUs working in tandem on high-performance computing simulations with AI. Combining the two is a golden bullet that propels us into a better future for understanding problems, Resch said.

Resch表示，一些客户需要更强大的AI处理能力，因此决定更新Hawk。加强版Hawk可以运行混合计算工作负载，里面的GPU和CPU可以在AI高性能计算模拟里串联工作。Resch称， 两者的组合是个利器，可以在美好的未来更好地推动各种问题的了解。

The University of Stuttgart is planning to use Hawk to train neural networks that can evaluate the quality and economic value of metal alloys, based on data from more than 2 billion simulations. It said the AI model it develops will be able to run on a standard personal computer and help companies that produce sheet metal to choose the best alloys for different products, such as a car door. This is extremely helpful in situations where experimentation is difficult or costly, Resch said.

斯图加特大学计划利用Hawk训练一些神经网络，用于评估金属合金的质量和经济价值，用到的数据基于来自超过20亿次模拟的数据。斯图加特大学表示，开发的AI模型可以在标准的个人计算机上运行，可以帮助生产金属板的公司在选择不同产品时找到最佳合金，例如汽车门一类的产品。Resch表示， 这对于那些做实验非常困难或非常昂贵情况非常有用。

Other projects earmarked for the beefed-up Hawk include an analysis of fluid dynamics data by a team of researchers from Stuttgart University. The idea is to train neural networks on data from simulations in order to create an AI model that can improve turbulence analysis. Tat will be useful for Airbus SE, the European aerospace corporation, which is collaborating with HLRS to mine its airflow data for better insights.

还有些专用拨款项目也用得上Hawk加强版，包括斯图加特大学研究人员团队的流体分析动力学数据分析项目。项目的基本想法是用模拟数据训练神经网络，最后可以创建能改善湍流分析的AI模型。这对于欧洲航空公司Airus SE非常有用。Airus SE与HLRS在合作挖掘旗下的气流数据，以获得更好的见解。

Another project will use AI to predict when hospital beds are likely to fill up in intensive care units from coronavirus cases, helping experts manage capacity more effectively. The goal of the project is to give policy-makers a four-week window to respond before maximum capacity is reached.

另一个项目则利用AI预测医院的重症监护病房由于冠状病毒患者而满员的预计时间点，进而帮助专家更有效地管理病床容量。该项目的目标是为政策制定者提供一个四周的窗口，以便在病床爆满前做出响应。

Its a critical question with so many people dying," Resch said. "Weve seen scenarios in places like Italy, New York and Wuhan where ICUs filled up in the first weeks of pandemic. So we will conduct simulations and predictions of the outlook for the pandemic over the next weeks and months, and GPUs will be extremely helpful for that.

Resch表示， 重症监护病房什么时候满员是个重要的问题，这么多的人濒临死亡。我们看到过意大利、纽约和武汉等地在大流行第一周的情景，重症监护病房第一周就爆满。我们因此会建个模拟系统预测未来几周及几个月内大流行的走向，新添的GPU对此非常有帮助。