由 英语数字智库 了解更多托福：自然科学综合词汇目录
适用人群：科技行业从业者、考研、雅思高阶学习者
外刊素材来源：
《MIT Technology Review（麻省理工科技评论）（美国科技期刊）》
报刊原版标题：
Computing Infrastructure and Chip Technology Iteration
一、英文改编正文
Advanced computing infrastructure serves as the fundamental backbone
supporting large-scale model training, real-time inference
and data-intensive artificial intelligence applications
across industrial and research sectors.
The rapid expansion of generative AI drives surging demand
for high-performance computing clusters,
which pushes chip designers to break
through traditional architectural limits
and rebalance computing power,
energy consumption and memory bandwidth.
Unlike general-purpose central
processing units optimized for versatile tasks,
graphics processing units and newly developed AI accelerators
are tailored for parallel matrix calculations,
the core mathematical operations required for neural network training.
Many semiconductor firms are exploring
advanced packaging technologies
to stack computing cores,
high-speed memory and signal transmission units
within a single compact chip package,
reducing data transmission latency
and cutting unnecessary power waste.
The continuous upgrade of chip manufacturing processes
enables smaller transistor sizes and higher transistor density,
yet the diminishing returns of miniaturization
force researchers to explore alternative
pathways beyond silicon-based circuits.
Regional computing hubs distribute computing tasks
between local edge devices and remote cloud data centers,
easing the burden on centralized server farms
and lowering the delay of data
processing for time-sensitive industrial scenarios.
Balanced development of computing hardware,
interconnection networks and cooling systems
determines the stable operation of super-scale computing facilities.
Policymakers and tech companies also
collaborate to formulate sustainable standards
for energy use and hardware recycling in the whole computing industry.
二、完整译文
先进算力基础设施是底层核心支柱，
支撑大模型训练、实时推理
以及工业与科研领域内
各类数据密集型人工智能应用。
生成式人工智能的快速扩张催生暴涨的算力需求，
推动高性能计算集群普及，
促使芯片设计者突破传统架构的局限，
在算力、能耗与内存带宽三者之间重新寻找平衡。
通用中央处理器为多类型任务而优化，与之不同，
图形处理器与全新研发的AI加速器，
专门针对并行矩阵运算做定制设计，
而矩阵运算正是神经网络训练所需的核心数学运算。
众多半导体企业探索先进封装技术，
将计算核心、高速内存与信号传输单元
堆叠在单个紧凑芯片封装内，
降低数据传输延迟，减少不必要的电力损耗。
芯片制造工艺的持续升级，
实现更小晶体管尺寸与更高晶体管密度，
但微型化带来的收益逐步递减，
迫使科研人员探索硅基电路之外的替代技术路线。
区域性算力枢纽在本地边缘设备与远端云数据中心之间
分配计算任务，
减轻中心化服务器集群的负载压力，
降低对时间敏感型工业场景的数据处理时延。
计算硬件、互联网络与散热系统的协同均衡发展，
决定超大规模算力设施能否稳定运行。
政策制定机构与科技企业也协同合作，
为整个算力行业制定能耗与硬件回收方面的可持续规范。
三、核心重点词汇短语
1. infrastructure
/ˈɪnfrəstrʌktʃə(r)/
基础设施
名词(n.)
2. iteration
/ˌɪtəˈreɪʃn/
迭代，更新
名词(n.)
3. inference
/ˈɪnfərəns/
推理，推断
名词(n.)
4. semiconductor
/ˌsemikənˈdʌktə(r)/
半导体
名词(n.)
5. latency
/ˈleɪtənsi/
延迟，时延
名词(n.)
6. transistor
/trænˈzɪstə(r)/
晶体管
名词(n.)
7. silicon
/ˈsɪlɪkən/
硅
名词(n.)
8. hub
/hʌb/
枢纽，中心
名词(n.)
四、长难句分层拆解
原句1
Advanced computing infrastructure serves as the fundamental backbone
supporting large-scale model training, real-time inference
and data-intensive artificial intelligence applications
across industrial and research sectors.
句子主干：
Infrastructure serves as backbone
分层修饰成分：
Advanced computing
多层前置定语，修饰基础设施
fundamental
前置定语，修饰支柱
supporting large-scale model training,
real-time inference and data-intensive artificial intelligence applications
现在分词短语作后置定语，说明基础设施支撑的业务
large-scale、real-time、data-intensive artificial intelligence
多层前置定语分别修饰训练、推理、应用
across industrial and research sectors
介词短语作地点状语，限定应用覆盖领域
原句2
Unlike general-purpose central
processing units optimized for versatile tasks,
graphics processing units and newly developed AI accelerators
are tailored for parallel matrix calculations,
the core mathematical operations required for neural network training.
句子主干：
GPUs and AI accelerators are tailored for calculations
分层修饰成分：
Unlike general-purpose central processing units optimized for versatile tasks
比较状语，和通用CPU做对比
general-purpose
前置定语，修饰中央处理器
optimized for versatile tasks
过去分词后置定语，修饰CPU
newly developed
前置定语，修饰AI加速器
parallel matrix
多层前置定语，修饰矩阵运算
the core mathematical operations
同位语，解释说明并行矩阵运算
required for neural network training
过去分词后置定语，修饰数学运算
原句3
Many semiconductor firms are exploring
advanced packaging technologies
to stack computing cores,
high-speed memory and signal transmission units
within a single compact chip package,
reducing data transmission latency
and cutting unnecessary power waste.
句子主干：
Firms are exploring technologies to stack components,
reducing latency and cutting waste
分层修饰成分：
Many semiconductor
多层前置定语，修饰企业
advanced packaging
多层前置定语，修饰技术
to stack computing cores, high-speed memory
and signal transmission units
不定式作目的状语，说明研发技术的目标
within a single compact chip package
地点状语，限定元件堆叠的空间
reducing data transmission latency
and cutting unnecessary power waste
现在分词短语作结果状语，描述这项技术带来的两项收益
data transmission、unnecessary power
多层前置定语修饰延迟、电力损耗
©英语数字智库 版权所有
所有学习素材资料均为英语数字智库原创，禁止私自转载、
摘抄、复制、二次编辑、商用分发，任何侵权行为将依法追究责任。
请点击返回首页麻省理工科技评论精选目录