VOA科学技术2023 什么是人工智能芯片？（在线收听）

What Are AI Chips?

什么是人工智能芯片？

Some kinds of computer chips have gained attention recently because they are used in computers linked to artificial intelligence (AI).

某些类型的计算机芯片最近引起了人们的关注，因为它们被用于与人工智能相关的计算机。

Some experts describe AI chips as similar to graphics chips that speed up complex video games. Makers of AI chips say they are designed for AI systems like ChatGPT and will make them faster and less costly.

一些专家形容人工智能芯片类似于加速复杂视频游戏的图形芯片。人工智能芯片制造商表示，它们是为ChatGPT等人工智能系统设计的，将使它们更快、成本更低。

The sudden interest in such chips is happening during what some experts call an AI revolution that could change the technology industry.

一些专家说人工智能革命可能改变科技行业，在此期间，人们突然对这种芯片产生了兴趣。

Nvidia, led by Jensen Huang, is a maker of graphics chips. The company is based in Santa Clara, California. Nvidia also is a designer of AI chips. Stock shares of the company increased in value by 25 percent last Thursday after company officials predicted a large increase in income. The company's stock was briefly valued at more than $1 trillion recently.

由黄仁勋领导的英伟达是一家图形芯片制造商。该公司总部位于加利福尼亚州圣克拉拉。英伟达也是人工智能芯片的设计者。上周四，公司官员预测收入将大幅增长，之后该公司股价上涨了25%。该公司的股票最近一度估值超过1万亿美元。

What are AI chips?

什么是人工智能芯片？

Hanna Dohmen is a researcher with the Center for Security and Emerging Technology in Washington, D.C. She said: "There really isn't a completely agreed upon definition of AI chips."

汉娜·多曼是位于华盛顿特区特区的安全与新兴技术中心的研究员。她说：“人们对人工智能芯片的定义还没有完全达成一致。”

The term is used to talk about computing equipment that is specialized to deal with AI workloads. For example, AI chips are used in "training" AI systems to work out difficult problems.

该术语用于谈论专门处理人工智能工作负载的计算设备。例如，人工智能芯片被用于“训练”人工智能系统来解决难题。

Coming from video games

出身于电子游戏

Three businessmen founded Nvidia in 1993 to expand the abilities of graphics on computers. Within a few years, the company had developed a new chip called a graphics processing unit, or GPU. GPUs speed up video games by performing several graphics jobs at the same time.

1993年，三位商人创立了英伟达，以扩展计算机的图形能力。几年内，该公司开发了一种名为图形处理单元的新芯片。图形处理单元通过同时执行多个图形任务来加速视频游戏。

The method is known as parallel processing. It would be important in the development of both games and AI.

这种方法被称为并行处理。它对游戏和人工智能的发展都很重要。

Two students from the University of Toronto used a GPU-based network to win a 2012 AI competition called ImageNet. The network, known as a neural network, could identify photo images at much lower error rates than competitors.

多伦多大学的两名学生使用基于图形处理单元的网络赢得了2012年ImageNet人工智能竞赛。该网络被称为神经网络，它可以以比竞争对手低得多的错误率识别影像制品。

The award began an interest in AI-related parallel processing. And it created new business for Nvidia and its competitors while providing researchers powerful tools for exploring AI development.

该奖项引发了人们对人工智能相关并行处理的兴趣。它为英伟达及其竞争对手开创出新的业务，同时为研究人员提供了探索人工智能开发的强大工具。

Modern AI chips

现代人工智能芯片

Eleven years after that ImageNet competition, Nvidia is the main supplier of chips for building and updating AI systems. One of its recent products, the H100 GPU, has 80 billion transistors. That is reportedly about 13 million more than Apple's latest top processor used in its MacBook Pro computers. However, this technology is costly. One online seller lists the H100 for $30,000.

在ImageNet竞赛举办11年后，英伟达成为构建和升级人工智能系统的芯片的主要供应商。其最新产品之一H100图形处理单元拥有800亿个晶体管。据报道，这比苹果在其MacBook Pro电脑中使用的最新顶级处理器多出约1300万个。然而，这项技术成本高昂。一位网络卖家将H100标价为3万美元。

Nvidia does not make these GPU chips itself. The company depends on manufacturers such as Taiwan Semiconductor Manufacturing and South Korea's Samsung Electronics.

英伟达自己并不生产这些图形处理单元芯片。该公司依赖中国台湾半导体制造公司和韩国三星电子等制造商。

Some of the biggest buyers for AI chips are cloud-computing services. Technology companies, like Amazon and Microsoft, use these chips for cloud computing.

云计算服务是人工智能芯片的最大买家之一。像亚马逊和微软这样的科技公司将这些芯片用于云计算。

These companies permit buyers to pay for temporary computing power to do AI jobs. Other companies then can use AI systems without a big investment in buildings and equipment. And the cloud-computer services include other tools for drug discovery or helping to oversee customers.

这些公司允许买家购买临时计算能力用于从事人工智能工作。其他公司则可以使用人工智能系统，而无需对建筑和设备进行大量投资。云计算机服务还包括其他用于新药研发或帮助监督客户的工具。

Other uses and competition

其他用途和竞争

Parallel processing has many uses outside of AI. A few years ago, for example, Nvidia graphics cards were in short supply because cryptocurrency miners had bought most of them. A cryptocurrency is a digital form of money that is not overseen or controlled by a government.

并行处理在人工智能之外有很多用途。例如，几年前，英伟达显卡供不应求，因为加密货币矿工购买了大部分显卡。加密货币是一种不受政府监管或控制的数字货币形式。

Cryptocurrency miners set up banks of computers to solve mathematical problems for cryptocurrency rewards. That problem went away as the cryptocurrency market collapsed in early 2022.

加密货币矿工建立了计算机群来解决加密货币奖励的数学问题。随着加密货币市场在2022年初崩溃，这个麻烦消失了。

Researchers say Nvidia will face more competition. One possible competitor is Advanced Micro Devices (AMD), which already competes with Nvidia in the market for computer graphics chips. AMD has recently taken steps to build its own lineup of AI chips.

研究人员表示，英伟达将面临更多的竞争。超微公司就是可能的竞争对手之一，该公司已经在计算机图形芯片市场上与英伟达展开竞争。超微公司最近已经着手建立自己的人工智能芯片阵容。