「英中对照」芯粒将如何改变服务器设计与数据中心

Chiplets let chipmakers build processors from multiple smaller dies rather than a single large monolithic die, providing greater flexibility in server CPU design.Getty Images芯粒（Chiplet）让芯片制造商能够用多个较小的裸片而非单个大尺寸单片裸片来构建处理器，从而为服务器 CPU 设计提供更大的灵活性。Getty Images

Since the advent of microprocessors more than 50 years ago, CPUs have typically been built as single pieces of silicon. As a result, servers and the data centers that house them have largely been designed around that model.自 50 多年前微处理器问世以来，CPU 通常被构建为单块硅片。因此，服务器以及承载它们的数据中心在很大程度上都是围绕这一模式设计的。

However, that is beginning to change. The growing adoption of chiplets – small modular dies that can be combined in a single processor package – is giving chipmakers a new way to build CPUs. By improving manufacturing flexibility and enabling more specialized processor designs, chiplets could reshape not just server architecture but also how data centers approach compute density, power efficiency, and infrastructure planning.然而，这种情况正在开始改变。芯粒——可以在单个处理器封装中组合的小型模块化裸片——的日益普及，为芯片制造商提供了一种构建 CPU 的新方式。通过提高制造灵活性并实现更专业化的处理器设计，芯粒不仅可能重塑服务器架构，还可能改变数据中心处理计算密度、能效和基础设施规划的方式。

What Are Chiplets?什么是芯粒（Chiplet）？

A chiplet is a small integrated circuit designed to perform a specific function within a larger processing package. Depending on the design, chiplets may handle compute, input/output (I/O), cache, memory connectivity, or workload-specific acceleration, such as AI acceleration.芯粒是一种小型集成电路，旨在更大的处理器封装内执行特定功能。根据设计的不同，芯粒可以负责计算、输入/输出（I/O）、缓存、内存连接或特定工作负载的加速任务，例如 AI 加速。

Related:AMD to Acquire World Labs for $8.2B to Advance AI Models and Robotics相关：AMD 将以 $8.2B 收购 World Labs，以推进 AI 模型与机器人技术

When combined, chiplets can provide the same overall capabilities as a traditional CPU. The key difference is that, instead of manufacturing a processor as one large monolithic die, chipmakers can build it from multiple smaller dies assembled in a single package.组合在一起后，芯粒可以提供与传统 CPU 相同的整体能力。关键区别在于，芯片制造商不再将处理器制造为单个大型单片裸片，而是可以用多个较小的裸片组装在单个封装中来构建处理器。

The chiplet concept has existed for decades, and related multi-die approaches are not entirely new. But adoption accelerated in the late 2010s as chipmakers sought more efficient ways to improve processor performance, manage manufacturing complexity, and scale designs across different workloads.芯粒概念已经存在了几十年，相关的多裸片方法也并非全新事物。但在 2010 年代后期，随着芯片制造商寻求更高效的方式来提升处理器性能、管理制造复杂性并跨不同工作负载扩展设计，其采用速度开始加快。

The Benefits of Chiplets芯粒的优势

A chiplet-based approach to CPU design offers several potential advantages.基于芯粒的 CPU 设计方法具有几个潜在优势。

Reduced manufacturing costs. Smaller dies are often easier to manufacture with higher yields than a single large, complex die. If a single die has a defect, the loss is usually less significant than losing an entire monolithic processor. It can also be easier to engineer and optimize a smaller die for a specific function than to design one large chip that must handle multiple functions at once.降低制造成本。与单个大型复杂裸片相比，较小的裸片通常更容易制造，良率也更高。如果单个裸片存在缺陷，损失通常小于损失整个单片处理器。此外，针对特定功能设计和优化较小的裸片，也可能比设计一个必须同时处理多种功能的大型芯片更容易。

Design freedom. By separating functions such as compute and I/O into different dies, chipmakers can tailor processor designs more efficiently for different types of workloads and product segments.设计自由度。通过将计算和 I/O 等功能分离到不同的裸片中，芯片制造商可以针对不同类型的工作负载和产品细分市场更高效地定制处理器设计。

Potentially higher compute density. In some cases, chiplets allow chipmakers to pack more processing capability into a single processor package, helping to enable more powerful servers without necessarily increasing physical server size.潜在的更高计算密度。在某些情况下，芯粒使芯片制造商能够将更多的处理能力集成到单个处理器封装中，从而在不增加服务器物理尺寸的情况下实现更强大的服务器。

Supply chain flexibility. Because smaller dies are often easier and faster to manufacture than very large monolithic chips, chiplet-based designs may reduce some production constraints. That said, they do not eliminate supply chain risk altogether, because advanced packaging and integration can create bottlenecks on their own.供应链灵活性。由于较小的裸片通常比非常大的单片芯片更容易、更快地制造，基于芯粒的设计可能会减少一些生产限制。尽管如此，它们并不能完全消除供应链风险，因为先进的封装和集成本身也可能造成瓶颈。

Related:GPU Lifespan in Data Centers: Physical vs. Economic相关：数据中心中的 GPU 寿命：物理寿命与经济寿命

In short, chiplets can make it easier for chipmakers to scale processor designs, improve yields, and optimize hardware for growing demand.简而言之，芯粒可以让芯片制造商更容易扩展处理器设计、提高良率，并优化硬件以满足不断增长的需求。

What Do Chiplets Mean for Data Centers?芯粒对数据中心意味着什么？

Although chiplets directly affect processor design, their impact could extend well beyond the chip itself.尽管芯粒直接影响处理器设计，但其影响可能远远超出芯片本身。

One important implication is the potential for increased compute density inside data centers. If chiplets allow server vendors to deliver more processing power within a single server platform, operators may be able to support heavier workloads without expanding physical footprints at the same pace.其中一个重要意义是数据中心内部计算密度有望提升。如果芯粒能让服务器厂商在单一服务器平台内提供更强的处理能力，运营商或许就能在不以同等速度扩大物理占地的情况下，支撑更繁重的工作负载。

Historically, higher compute density has often meant higher power draw and greater cooling requirements. Chiplets do not automatically solve that problem. However, if they enable more efficient processor designs or better performance per watt, they could help offset some of the infrastructure strain that typically comes with denser compute environments.从历史上看，更高的计算密度往往意味着更高的功耗和更大的散热需求。芯粒并不能自动解决这一问题。然而，如果它们能带来更高效的处理器设计或更高的每瓦性能，就可能有助于缓解更高密度计算环境通常带来的部分基础设施压力。

A key result of greater chiplet adoption, then, could be data centers that support higher IT loads without requiring proportionally more space or power. In that scenario, existing facilities could become more valuable because operators would be able to do more with the infrastructure they already have.因此，芯粒更广泛采用的一个关键结果可能是：数据中心能够支撑更高的 IT 负载，而无需按比例增加空间或电力。在这种情况下，现有设施将变得更有价值，因为运营商能够利用已有的基础设施做更多事情。

Related:‘Out of Hyperbole’: Nvidia’s AI Boom Tests Data Center Infrastructure Limits相关阅读：《抛开炒作：英伟达的 AI 热潮考验数据中心基础设施极限》

Chiplets may also help reduce the risk of server procurement delays by giving manufacturers more flexibility in how processors are designed and produced. This does not remove supply chain challenges altogether, but it could make sourcing CPUs and servers somewhat easier over time.芯粒还可以通过让制造商在处理器的设计和生产方式上拥有更大灵活性，帮助降低服务器采购延迟的风险。这并不能完全消除供应链挑战，但随着时间推移，它可能会让 CPU 和服务器的采购变得相对容易一些。

In the longer term, chiplets could also encourage more specialized server designs. As chipmakers gain flexibility to mix and match dies optimized for different functions, server platforms may become increasingly tailored for workloads such as AI, analytics, storage, or high-performance computing. That would give data center operators more options, but it could also make infrastructure planning more workload-specific than it is today.从长远来看，芯粒还可能推动更加专业化的服务器设计。随着芯片制造商获得灵活组合针对不同功能优化的小芯片（die）的能力，服务器平台可能会越来越多地为 AI、分析、存储或高性能计算等工作负载量身定制。这将给数据中心运营商提供更多选择，但也可能使基础设施规划比今天更加针对特定工作负载。

A Shift Worth Watching一个值得关注的变化

Chiplets are no longer a futuristic concept. They are already becoming an important part of modern processor design, especially as chipmakers look for new ways to improve performance and efficiency.芯粒不再是未来概念。它们已经成为现代处理器设计的重要组成部分，尤其是在芯片制造商不断寻找提升性能和效率新途径的背景下。

For data centers, the biggest impact may not come from chiplets themselves, but from the servers they enable. If chiplet-based designs continue to improve compute density, efficiency, and workload specialization, they could play a major role in shaping the next generation of data center infrastructure.对数据中心而言，最大的影响可能并非来自芯粒本身，而是来自由它们催生的服务器。如果基于芯粒的设计持续提升计算密度、能效和工作负载专业化程度，它们就可能在塑造下一代数据中心基础设施方面发挥重要作用。

About the Author关于作者

Technology Analyst技术分析师

Christopher Tozzi is a technology analyst with subject matter expertise in cloud computing, application development, open source software, virtualization, containers and more. He also lectures at a major university in the Albany, New York, area. His book, “For Fun and Profit: A History of the Free and Open Source Software Revolution,” was published by MIT Press.Christopher Tozzi 是一位技术分析师，在云计算、应用开发、开源软件、虚拟化、容器等领域拥有专业知识。他还在纽约州奥尔巴尼地区的一所重点大学授课。他的著作《For Fun and Profit: A History of the Free and Open Source Software Revolution》由 MIT Press 出版。