『内存墙』指的是处理器与内存之间怎样的矛盾？
What contradiction between processor and memory does the 'memory wall' describe?
几十年来 CPU 速度按摩尔定律飞涨，而内存访问延迟下降缓慢，处理器常因等数据而空转，性能被带宽和延迟卡住。Wulf 与 McKee 在 1995 年提出该术语，并推动多级缓存、预取等技术来缓解。
CPU speed rose fast with Moore's law while memory latency improved slowly, so processors often stalled waiting for data and hit a bandwidth/latency ceiling. Wulf and McKee coined the term in 1995 and spurred caches, prefetching, and other mitigations.
难度 专业 · 适合 高中(15-18)
CC BY 4.0 · 已核验 · Verified 2026-11-03
来源 · Source
Wulf 与 McKee《撞上内存墙》(ACM SIGARCH, 1995)；Hennessy 与 Patterson《计算机体系结构》
Wulf & McKee, 'Hitting the Memory Wall' (ACM SIGARCH, 1995); Hennessy & Patterson, 'Computer Architecture'
独立核验依据 · Independent verification: 核验依据 1 ISBN 9787111364580（豆瓣书目条目: 计算机体系结构）；本条出处含多部著作，此标识只担保上面点名
#内存墙 #缓存 #访存延迟 #带宽
GPU 相比 CPU 为什么特别擅长并行运算？
Why are GPUs so much better than CPUs at parallel computation?
GPU 内部有成千上万个轻量核心，专为大量相同操作（如矩阵乘法）的『单指令多数据』并行而设计，适合图形渲染和深度学习。NVIDIA 的 CUDA 编程模型让程序员能把通用计算映射到这些核心。
A GPU packs thousands of lightweight cores tuned for 'single-instruction-multiple-data' parallelism on uniform operations like matrix multiply, ideal for graphics and deep learning. NVIDIA's CUDA model lets programmers map general computing onto those cores.
难度 专业 · 适合 高中(15-18)
CC BY 4.0 · 已核验 · Verified 2026-11-04
来源 · Source
Nickolls 等《NVIDIA CUDA 编程模型》(IEEE Micro, 2008)；Owens 等,《GPU Computing》,Proceedings of the IEEE,2008 (DOI: 10.1109/jproc.2008.917757)
Nickolls et al., 'NVIDIA CUDA programming model' (IEEE Micro, 2008); Owens et al., 'GPU Computing', Proceedings of the IEEE, 2008, DOI: 10.1109/jproc.2008.917757
独立核验依据 · Independent verification: 核验依据 1 Crossref DOI 10.1109/jproc.2008.917757
#GPU #并行 #CUDA #SIMD
指令流水线是怎样让 CPU 同时『处理多条指令』的？
How does instruction pipelining let a CPU work on several instructions at once?
流水线把一条指令的执行拆成取指、译码、执行、写回等阶段，不同指令的不同阶段在时间上重叠，像工厂装配线一样提升吞吐。代价是指令相关和分支会带来停顿或冲刷。
Pipelining splits an instruction into stages—fetch, decode, execute, write-back—so different instructions occupy different stages at the same time, raising throughput like a factory line. The cost is stalls or flushes from dependencies and branches.
难度 专业 · 适合 高中(15-18)
CC BY 4.0 · 已核验 · Verified 2026-10-04
来源 · Source
Hennessy 与 Patterson《计算机体系结构：量化研究方法》；Patterson 与 Hennessy《计算机组成与设计》
Hennessy & Patterson, 'Computer Architecture: A Quantitative Approach'; Patterson & Hennessy, 'Computer Organization and Design'
独立核验依据 · Independent verification: 核验依据 1 ISBN 9781558604285 (2009)
#流水线 #指令级并行 #停顿 #吞吐
RISC 与 CISC 两种指令集哲学的根本分歧是什么？
What is the fundamental split between RISC and CISC instruction-set philosophies?
CISC 倾向用较少、较复杂、功能繁多的指令；RISC 则精简为少量简单、定长、单周期为主的指令，把复杂工作交给编译器与流水线。Patterson 等在 1980 年代的系统研究使 RISC 成为 ARM、RISC-V 等现代设计的主流。
CISC favors fewer, more complex, feature-rich instructions; RISC pares down to a small set of simple, mostly fixed-length, fast instructions, leaving complexity to the compiler and pipeline. Patterson's 1980s work made RISC dominant in ARM, RISC-V, and modern designs.
难度 专业 · 适合 高中(15-18)
CC BY 4.0 · 已核验 · Verified 2026-11-03
来源 · Source
Patterson《精简指令集计算机》(1981)；Hennessy 与 Patterson《计算机体系结构》
Patterson, 'Reduced Instruction Set Computers' (1981); Hennessy & Patterson, 'Computer Architecture'
独立核验依据 · Independent verification: 核验依据 1 OPENLIBRARY 标识 /works/OL2664271W：Computer architecture（首
#RISC #CISC #指令集 #编译器
冯·诺依曼体系结构最核心的思想是什么？
What is the central idea of the von Neumann architecture?
它把指令和数据存放在同一可读写的存储器中，由控制器逐条取指、译码、执行，形成『存储程序』范式。这一思想写入 1945 年的《第一份草案》，至今仍是绝大多数计算机的基础。
It stores instructions and data in the same read-write memory and has a control unit fetch, decode, and execute them one by one—the 'stored-program' paradigm. Laid out in the 1945 'First Draft', it still underlies most computers.
难度 进阶 · 适合 高中(15-18)
CC BY 4.0 · 已核验 · Verified 2026-11-02
来源 · Source
von Neumann《关于 EDVAC 的报告草案》(1945)；Hennessy 与 Patterson《计算机体系结构：量化研究方法》
von Neumann, 'First Draft of a Report on EDVAC' (1945); Hennessy & Patterson, 'Computer Architecture: A Quantitative Approach'
独立核验依据 · Independent verification: 核验依据 1 CROSSREF 标识 10.5479/sil.538961.39088011475779：First draf
#冯诺依曼 #存储程序 #指令 #存储器