How does computer hardware work | 现代数学启蒙

slug

type

status

无字幕版本：

有字幕版本：

Test :

https://zvjqzp.form.top/f/wZqoMH

Vocabulary：

Desktop computer - 台式电脑

Dissection - 解剖

3D animation - 三维动画

Hardware - 硬件

Microscope - 显微镜

Nanoscopic view - 纳米级视图

Transistor - 晶体管

Integrated circuit (IC) - 集成电路

Die - 芯片

Printed circuit board (PCB) - 印刷电路板

Central Processing Unit (CPU) - 中央处理器

Integrated heat spreader - 集成散热片

Landing grid array - 焊盘阵列

Motherboard - 主板

Core - 核心

L3 memory cache - L3缓存

Ring interconnect - 环形互连

Integrated graphics processor - 集成图形处理器

Memory controller - 内存控制器

System agent - 系统代理

Platform I/O - 平台输入/输出

Chipset - 芯片组

Display - 显示器

Solid State Drive (SSD) - 固态硬盘

M2 slot - M2插槽

Ethernet - 以太网

Wi-Fi - 无线网

SATA port - SATA端口

PCIe slot - PCIe插槽

Voltage regulator module (VRM) - 电压调节模块

Power supply - 电源

Transformer - 变压器

Opto-isolator - 光电耦合器

Switching power transistor - 开关电源晶体管

Filter - 过滤器

Graphics Processing Unit (GPU) - 图形处理器

HDMI - 高清多媒体接口

Display Port - 显示端口

Streaming multiprocessor - 流多处理器

Floating-point arithmetic - 浮点运算

DRAM - 动态随机存取存储器

Memory bank - 存储库

1T1C memory cell - 1T1C存储单元

3D NAND - 3D NAND闪存

Charge trap - 电荷陷阱

NVMe - 非易失性内存主机控制器接口

SATA - 串行ATA

Hard disk drive (HDD) - 硬盘驱动器

Spindle - 主轴

Read/write head - 读/写磁头

Magnetic domain - 磁畴

Scroll wheel - 滚轮

Infrared light - 红外线

Image sensor - 图像传感器

Battery - 电池

Processor - 处理器

Keyboard - 键盘

Plastic traces - 塑料电路

Engineering education - 工程教育

Throughout this video we’re going to do a desktop computer dissection using 3D animation. Kind of like a dissection lab in biology class, but instead, we’ll journey through the inside of this computer and disassemble every piece of [hardware ]. We’ll then use a microscope and zoom in to give you a nanoscopic view of the [transistors ] and other structures inside. To make this video we disassembled all the [hardware ] inside a typical desktop computer, we desoldered and removed the components from each of the [printed circuit boards ] and took thousands of pictures. Then using these pictures, we meticulously 3D modeled every single component from the [computer case ] down to the tiniest [resistor ]. As a result, here’s all the [hardware ] that we’ll explore throughout this video. It kind of looks like a crime scene where someone viciously destroyed a computer and arranged all the components, but anyways let’s dive right in.

We’ll begin this journey with the [Central Processing Unit ] or [CPU ] which is the brain of the computer. On top we have the cover, called an [integrated heat spreader ], and inside is a smaller metal package that holds the [integrated circuit ] which is technically called a [die ]. This [die ] is mounted on a [printed circuit board ] that distributes the 1200 connection points to landing pads that interface with the [landing grid array ] on the [motherboard ]. The [integrated circuit ] inside has a few different sections, but perhaps the most recognizable are the 10 cores where programs and instructions are run. Each core is quite complicated, so here’s a diagram laying out the different functional sections. There are dozens of rather complicated elements in this diagram, and you can look forward to a series of videos we’re currently planning which will explore [computer architecture ] and how each of these sections work.

Let’s zoom into a nanoscopic view of the [integrated circuit ] so we can see individual [transistors ]. These [transistors ] are incredibly small, only a few nanometers wide and in this [die ] there are approximately 8 to 10 billion. On top of the [transistors ] are multiple layers of metal wires with vias rising vertically between the layers. Together these [transistors ] and wires create a multilayer labyrinth or highway resulting in a computer that can execute billions of operations every second. Let’s now zoom out and look at other sections of the [CPU ]. To the side of each core is the shared [L3 memory cache ] and [ring interconnect ]. On the far right is an [integrated graphics processor ] which functions as a less powerful [GPU ]; in the top left is the [memory controller ] which sends data to and from the [DRAM ]; and finally on the far left is the [system agent ] and [platform I/O ], which communicates with the [Chipset ] on the [motherboard ] and manages the flow of data between many of the other components in your PC.

This [motherboard ] is a massive, [printed circuit board ] with thousands of wires running inside and a variety of [microchips ], components, sockets, ports, slots, headers, and connectors soldered to it. Perhaps the single most important and expensive component, aside from the bare PCB, is the [Chipset ], which is this [integrated circuit ] found underneath a heat sink down here and is connected directly to the [system agent ] section in the [CPU ]. Here’s a diagram illustrating how the [CPU ] and [chipset ] are connected to everything else. As seen here, the [CPU ] connects directly to the [DRAM ], one or two displays, the [GPU ], and perhaps a few [SSDs ] plugged into the [M2 slots ]. The [Chipset ] manages most everything else: data flowing through the [ethernet ] or [Wi-Fi ], data going to and from [solid state drives ] and [hard drives ] plugged into the [SATA ports ], some of the [PCIe slots ], your keyboard and mouse, USB devices, and the audio sent to the speakers or from the microphone.

Just a quick note, [computer hardware ] has evolved immensely over the past 65 years and continues to evolve, so the details we show should be thought of as a current day example PC and not as how all computers work. Let’s move on and skip over the many different sockets and connectors throughout the [motherboard ] and focus on the [voltage regulator module ] or [VRM ] found near the [CPU ]. These components are used to drop the voltage coming from the [power supply ] down to the 1.3 volts used by the [CPU ]. As a result of all the power flowing through these components and their 80 to 90 percent efficiency, heat sinks need to be placed on top. While we’re on the topic of power, this [CPU ] consumes power equivalent to approximately 16 LED bulbs, thus generating a considerable amount of heat which is taken away by the [CPU cooler ]. This particular cooler uses a pump to circulate liquid through these tubes and into the radiator’s channels which transfers heat to the radiator fins. The fans then help transfer the heat to the air and the cooled liquid returns to the pump via the return loop. The pump is a brushless [DC motor ] constructed from a [PCB ], a control chip, and a stator on the dry side, a barrier in the middle, and then the permanent magnet rotor and impeller on the liquid side. There’s no mechanical connection between the rotor and stator, thereby preventing any leaks of the cooling liquid.

Let’s move onto the [power supply ] which distributes power throughout the computer. In here, the main transformer reduces the voltage and bridges the isolation boundary between the primary side high voltage and the secondary side lower voltages used throughout your computer. Here’s the control [PCB ] that ensures a stable output voltage and sends adjustment signals to the switching power transistor on the primary side using opto-isolators. There are dozens of other components used to filter the input voltage and generate various output voltages which are then sent to all the different [hardware ] in your computer. For example, an [SSD ] consumes just a few watts, and the connector uses these voltages, whereas your [GPU ] can consume hundreds of watts using these connectors and these voltages.

Next, we’ll explore the [GPU ], but before that, let’s take a step back and consider the technology we’ve covered thus far. Alone, each of these components doesn’t do much at all, but when united, they combine to form a powerful system. Similarly, this video is a multidisciplinary combination of engineering, technology, art, and animation inspired by the Magic School Bus.

Let’s get back to dissecting this computer and move onto the [graphics card ] and [GPU ] which is also the brain of the computer… well, actually this analogy doesn’t really work very well but anyway… Opening the [graphics card ] we see another [PCB ], with the [GPU’s integrated circuit ] in the center, [VRAM ] chips all around, and the [voltage regulator module ] on the side. Above the IC is a heat sink with a fan to dissipate heat, and on the side of the [graphics card ] are [HDMI ] and [Display Ports ], a [PCIe Interface ], and then on the other side is the input power connector.

Let’s focus on the [GPU integrated circuit ] which, similar to the [CPU ], has more than a thousand solder pads that connect it to the [PCB ]. Opening the packaging we find the [GPU die ] which is noticeably different from the [CPU ]. In here are approximately 11.8 billion [transistors ] organized into 6 [graphics processing clusters ], totaling 28 [streaming multiprocessors ]. Each [streaming multiprocessor ] is composed of 128 cores, resulting in a total of 3,584 cores. Each core has sections for integer and floating-point arithmetic, and sections for queueing in the operands and collecting the results and is far simpler than a [CPU core ]. Additionally, on the [die ] is an [L2 memory cache ] shared among all the [graphics processing clusters ], a set of [memory controllers ] that connect to the [VRAM ] located around the processor, and a [PCIe interface ] for connecting to the [CPU ].

When we zoom in to see a nanoscopic view of the [integrated circuit ] we find something very similar to what we saw in the [CPU ], with the [transistors ] on the bottom and a labyrinth of multiple layers of metal wires above. All these structures are manufactured in multibillion dollar semiconductor fabrication plants or fabs but that’s a topic for a whole different video, so let’s move on. [GPUs ] and [CPUs ] are similar in many ways, however, [GPUs ] have thousands of cores that are limited to basic arithmetic, whereas [CPUs ] have only a handful of cores that perform far more complicated operations. Additionally, [CPUs ] have branch prediction and deep pipelines that optimize the execution of code. Here’s a quick example of what [GPUs ] can do. Take this image comprised of 16 million pixels each with [RGB values ] for each pixel. A simple way to brighten the image is to add 20 to each of these numbers. A [CPU ] has 10 cores and thus performs arithmetic 10 numbers at a time whereas [GPUs ] distribute the data to thousands of cores thus performing magnitudes more parallel processing.

Let’s close this [graphics card ] and talk briefly about the 3D model. Note that this is a slightly older model [graphics card ] because we buy most of our [hardware ] as non-functioning parts from eBay, and then, as shown earlier, we tear them down rather destructively in order to accurately model everything.

Additionally, sometimes components shown aren’t compatible, such as this [motherboard ] that says DDR4, while the [DRAM ] is DDR5.

That said, modeling and animating everything to make it feel like you’re actually inside a computer took us around 500 hours and we would greatly appreciate it if you could take a few seconds to hit that like button, subscribe if you haven’t already, type up a quick comment below, and share this video with someone who will enjoy it. Also, we have a Patreon and would appreciate any support. We’re planning more videos on [computer architecture ] and other related topics and can’t do it without your help, so thank you for doing these 4 quick things. It helps a ton.

Let’s move on and look at the [DRAM ], [Solid State Drives ], and [Hard Drives ]. We’re not going to spend too long, because we have an entire series of videos on [solid state drives ], and then separate videos covering [DRAM ], and [Hard Drives ]. But quickly, the [CPU ] communicates directly with the [DRAM ] through memory channels running inside the [motherboard ]. Inside each of these 8 [DRAM chips ] is an [integrated circuit ] composed of 32 [memory banks ] each 8192 columns wide by 65536 rows tall. The [DRAM ] temporarily stores data using capacitors and [transistors ] called 1T1C memory cells in 2D arrays that look like this. Data can be accessed within nanoseconds, however, among these 8 chips, only 16 Gigabytes of data can be temporarily stored. Take a look at our 35-minute video on [DRAM ], but for now let’s move on to [SSDs ] which permanently store data in 3D arrays called 3D [NAND ]. This array is 100 to 200 layers tall, 32 to 64 thousand columns wide and 32 to 64 thousand rows deep. Additionally, within a single [SSD chip ] such as this one, are multiple 3D [NAND arrays ] stacked one on top of the other. As a result, a single [microchip ] can store terabytes of data, however reading or writing data takes 50 or so microseconds, which is 3000 times slower than [DRAM ]. Zooming in on a single [SSD memory cell ] we find a charge trap which stores different levels of charge allowing for 3 bits of data to be more permanently saved. Looking at the [NVMe ] and [SATA SSD ], both have a few 3D [NAND data storage chips ], a [DRAM chip ] for buffering and holding the data mapping table and a [controller chip ].

Let’s move on and dissect the [hard disk drive ]. Here we have a disk mounted to a spindle with a motor that rotates the disk at thousands of rpm. A read write head moves across the disk in order to access a single track out of half a million other data tracks. Let’s zoom in on the read write head. The write head changes the direction of localized magnetic domains in a small layer on the disk whereas the read head senses these changes in magnetic domains. This [disk drive ] is even slower than the [SSD ], taking a few milliseconds to access, thus resulting in slower read and write times, but costing less per terabyte of storage.

Thus far we’ve covered all the [hardware ] inside your computer. So, thanks for watching this far, and as a bonus here’s what it looks like inside a [computer mouse ], with the scroll wheel up top, the infrared light, image sensor, and multiple lenses down here and the battery and processor in the middle. For [computer mice ] we have separate dedicated videos exploring the image sensor and scroll wheel with incredible details. Additionally, here’s what it looks like inside a basic [keyboard ], with plastic traces that carry electricity to each key and, when pressed, that key completes a circuit which is sensed by the processor up here.

That’s pretty much it for what it looks like inside your computer. We believe the future will require a strong emphasis on engineering education and we’re thankful to all our Patreon and YouTube Membership Sponsors for supporting this dream. If you want to support us on YouTube Memberships, or Patreon, you can find the links in the description. Also, thank you again to Brilliant for sponsoring this video.

This is Branch Education, and we create 3D animations that dive deep into the technology that drives our modern world. Watch another Branch video by clicking one of these cards or click here to subscribe. Thanks for watching to the end!

在这个视频中，我们将通过3D动画进行台式电脑解剖。这有点像生物课上的解剖实验室，但我们将探索这台电脑的内部并拆解每个硬件。然后，我们将使用显微镜放大，给你一个晶体管和其他结构的纳米级视角。为了制作这个视频，我们拆解了一个典型台式电脑的所有硬件，从每个印刷电路板上拆焊并移除了组件，并拍摄了数千张照片。然后，使用这些照片，我们精心3D建模了从电脑机箱到最小电阻器的每一个组件。因此，在整个视频中，我们将探索所有这些硬件。看起来有点像某人恶意破坏了一台电脑并排列了所有组件的犯罪现场，但不管怎样，让我们直接深入探讨。

我们将从中央处理器（CPU）开始，这是电脑的大脑。在顶部，我们有一个称为集成散热器的盖子，里面是一个更小的金属封装，包含技术上称为芯片的集成电路。这个芯片安装在一个印刷电路板上，分布着1200个连接点，与主板上的栅格阵列接口。集成电路内部有几个不同的部分，但也许最显眼的是10个运行程序和指令的核心。每个核心都相当复杂，所以这里有一个图表展示了不同的功能部分。这个图表中有几十个相当复杂的元素，你可以期待我们目前正在计划的一系列视频，这些视频将探讨计算机架构以及这些部分的工作原理。

让我们放大到集成电路的纳米级视角，以便看到单个晶体管。这些晶体管非常小，仅几纳米宽，在这个芯片中大约有80到100亿个。在晶体管之上有多层金属线，垂直贯穿各层的导孔。晶体管和金属线共同组成了一个多层迷宫或高速公路，使电脑能够每秒执行数十亿次操作。现在让我们缩小视角，看看CPU的其他部分。在每个核心的旁边是共享的L3缓存和环形互连。在最右边是一个集成图形处理器，功能上类似于一个不太强大的GPU；在左上角是内存控制器，负责发送和接收DRAM的数据；最后，在最左边是系统代理和平台I/O，负责与主板上的芯片组通信，并管理PC中许多其他组件之间的数据流。

主板是一个庞大的印刷电路板，内部有数千条电线，并焊接有各种微芯片、组件、插座、端口、插槽、引脚和连接器。也许除了裸PCB之外，最重要和最昂贵的组件是芯片组，这是一个集成电路，位于这里的散热片下方，并直接连接到CPU中的系统代理部分。这里有一个图表展示了CPU和芯片组如何与其他所有东西连接。正如这里所见，CPU直接连接到DRAM、一到两个显示器、GPU，或许还有插入M2插槽的几个SSD。芯片组管理几乎所有其他东西：通过以太网或Wi-Fi传输的数据，流向和来自插入SATA端口的固态硬盘和机械硬盘的数据，一些PCIe插槽，你的键盘和鼠标，USB设备，以及发送到扬声器或来自麦克风的音频。

需要注意的是，计算机硬件在过去65年里进化得非常快，并且还在继续进化，所以我们展示的细节应该被视为当前的例子，而不是所有计算机的工作方式。让我们继续跳过主板上许多不同的插座和连接器，聚焦于CPU附近的电压调节模块（VRM）。这些组件用于将来自电源的电压降低到CPU使用的1.3伏。由于这些组件中流过的大量电能以及它们80到90%的效率，散热片需要放置在顶部。说到电源，这个CPU消耗的电能相当于大约16个LED灯泡，因此产生了大量热量，这些热量通过CPU散热器被带走。这个特定的散热器使用泵通过这些管子循环液体，进入散热器的通道，将热量传递到散热器鳍片。风扇帮助将热量传递到空气中，冷却的液体通过回流环返回泵。泵是一个无刷直流电机，由一个PCB、一个控制芯片和干侧的定子构成，中间有一个隔离层，然后是液体侧的永磁转子和叶轮。转子和定子之间没有机械连接，从而防止冷却液泄漏。

接下来让我们讨论电源，它将电能分配到整个计算机。在这里，主变压器降低电压，并桥接初级高压侧和整个计算机中使用的次级低压侧之间的隔离边界。这里是控制PCB，确保稳定的输出电压，并使用光隔离器将调整信号发送到初级侧的开关电源晶体管。还有几十个其他组件用于过滤输入电压，并生成各种输出电压，然后发送到计算机中的所有不同硬件。例如，一个SSD仅消耗几瓦，而连接器使用这些电压，而你的GPU可能使用数百瓦，通过这些连接器和这些电压。

接下来我们将探讨GPU，但在此之前，让我们回顾一下到目前为止介绍的技术。单独来看，这些组件中的每一个都几乎不起作用，但当它们联合在一起时，它们形成了一个强大的系统。同样，这个视频是工程、技术、艺术和动画的跨学科结合，灵感来自《神奇校车》。

让我们回到解剖这台电脑，继续探讨显卡和GPU，这也是电脑的大脑……好吧，其实这个类比并不太合适，不管怎样……打开显卡，我们看到另一个PCB，中央是GPU的集成电路，周围是显存芯片，旁边是电压调节模块。IC上方是一个带风扇的散热片，用于散热，显卡侧面是HDMI和显示端口、PCIe接口，另一侧是输入电源连接器。

让我们聚焦于GPU集成电路，类似于CPU，它有超过一千个焊点将其连接到PCB。打开封装，我们发现GPU芯片与CPU明显不同。这里有大约118亿个晶体管，组织成6个图形处理簇，总共28个流处理器。每个流处理器由128个核心组成，总共3584个核心。每个核心有整数和浮点算术部分，以及用于排队操作数和收集结果的部分，比CPU核心简单得多。此外，芯片上还有所有图形处理簇共享的L2缓存、一组连接到显存的内存控制器，以及用于连接CPU的PCIe接口。

当我们放大到集成电路的纳米级视角时，发现的结构与在CPU中看到的非常相似，底部是晶体管，上方是多层金属线的迷宫。所有这些结构都是在价值数十亿的半导体制造厂或工厂中制造的，但这是一个完全不同的视频话题，所以让我们继续。GPU和CPU在许多方面相似，但GPU有数千个核心，这些核心仅限于基本的算术操作，而CPU只有少量核心，执行复杂得多的操作。此外，CPU具有分支预测和深度流水线，优化了代码的执行。这里有一个GPU能做什么的简单例子。这个图像由1600万个像素组成，每个像素都有RGB值。一个简单的方式来提高图像亮度是将每个数值增加20。CPU有10个

核心，因此一次执行10个数值的算术运算，而GPU将数据分配给数千个核心，从而执行数量级更多的并行处理。

让我们关闭显卡，并简要谈谈3D模型。请注意，这是一款稍旧的显卡模型，因为我们从eBay购买大多数硬件作为不工作的部件，然后，如前所示，我们对它们进行破坏性拆解，以准确建模。此外，有时显示的组件并不兼容，比如这个主板标示DDR4，而DRAM是DDR5。

也就是说，建模和动画化所有内容，使其感觉像你真的在电脑内部花费了我们大约500小时，我们非常感激你能花几秒钟点击赞按钮，订阅如果你还没有订阅，写一个简短的评论，并与会喜欢这视频的人分享。此外，我们有一个Patreon，非常感谢任何支持。我们正在计划更多关于计算机架构和其他相关主题的视频，离不开你的帮助，所以感谢你做这四件事。它非常有帮助。

让我们继续看看DRAM、固态硬盘和机械硬盘。我们不会花太多时间，因为我们有一整系列关于固态硬盘的视频，然后是单独的视频介绍DRAM和机械硬盘。但简单来说，CPU通过内存通道直接与DRAM通信，运行在主板内部。每个这些8个DRAM芯片内部有一个集成电路，由32个内存银行组成，每个银行有8192列，65536行。DRAM通过称为1T1C存储单元的电容和晶体管以2D阵列临时存储数据，看起来是这样的。数据可以在纳秒内访问，然而，在这8个芯片中，只能临时存储16GB的数据。观看我们关于DRAM的35分钟视频，但现在让我们继续讨论SSD，它们以3D阵列永久存储数据，称为3D NAND。这个阵列有100到200层高，32到64000列宽，32到64000行深。此外，在单个SSD芯片中，如此一个，多个3D NAND阵列叠加在一起。结果，一个微芯片可以存储TB的数据，然而，读取或写入数据需要大约50微秒，比DRAM慢3000倍。放大单个SSD存储单元，我们发现一个电荷捕获器，存储不同的电荷水平，允许永久保存3位数据。看一下NVMe和SATA SSD，两个都有几个3D NAND数据存储芯片，一个DRAM芯片用于缓存和保存数据映射表，还有一个控制芯片。

让我们继续解剖机械硬盘。这里我们有一个安装在主轴上的磁盘，电机以每分钟数千转的速度旋转磁盘。读写头在磁盘上移动，以访问五十万个数据轨道中的一个。让我们放大读写头。写头改变磁盘小层中的局部磁域方向，而读头感应这些磁域变化。这个磁盘驱动器甚至比SSD更慢，访问时间需要几毫秒，因此读取和写入时间更慢，但每TB的存储成本更低。

到目前为止，我们已经介绍了计算机内部的所有硬件。所以，感谢你观看到这里，作为奖励，这里是计算机鼠标内部的样子，顶部是滚轮，红外线灯，图像传感器和多个镜头在这里，中间是电池和处理器。我们有单独的视频详细介绍计算机鼠标的图像传感器和滚轮。此外，这里是基本键盘内部的样子，塑料电路传导电流到每个键，当按下时，该键完成电路，由这里的处理器感应。

这几乎就是你的电脑内部的样子。我们认为未来需要强调工程教育，非常感谢所有的Patreon和YouTube会员赞助者支持这个梦想。如果你想在YouTube会员或Patreon上支持我们，可以在描述中找到链接。再次感谢Brilliant赞助这个视频。

这是Branch Education，我们创建3D动画，深入探讨驱动现代世界的技术。点击这些卡片观看另一部Branch视频，或点击这里订阅。感谢观看到最后！

Throughout this video we’re going to do a desktop computer dissection using 3D animation. Kind of like a dissection lab in biology class, but instead, we’ll journey through the inside of this computer and disassemble every piece of hardware. We’ll then use a microscope and zoom in to give you a nanoscopic view of the transistors and other structures inside. To make this video we disassembled all the hardware inside a typical desktop computer, we desoldered and removed the components from each of the printed circuit boards and took thousands of pictures. Then using these pictures, we meticulously 3D modeled every single component from the computer case down to the tiniest resistor. As a result, here’s all the hardware that we’ll explore throughout this video. It kind of looks like a crime scene where someone viciously destroyed a computer and arranged all the components, but anyways let’s dive right in.

We’ll begin this journey with the Central Processing Unit or CPU which is the brain of the computer. On top we have the cover, called an integrated heat spreader, and inside is a smaller metal package that holds the integrated circuit which is technically called a die. This die is mounted on a printed circuit board that distributes the 1200 connection points to landing pads that interface with the landing grid array on the motherboard. The integrated circuit inside has a few different sections, but perhaps the most recognizable are the 10 cores where programs and instructions are run. Each core is quite complicated, so here’s a diagram laying out the different functional sections. There are dozens of rather complicated elements in this diagram, and you can look forward to a series of videos we’re currently planning which will explore computer architecture and how each of these sections work.

Let’s zoom into a nanoscopic view of the integrated circuit so we can see individual transistors. These transistors are incredibly small, only a few nanometers wide and in this die there are approximately 8 to 10 billion. On top of the transistors are multiple layers of metal wires with vias rising vertically between the layers. Together these transistors and wires create a multilayer labyrinth or highway resulting in a computer that can execute billions of operations every second. Let’s now zoom out and look at other sections of the CPU. To the side of each core is the shared L3 memory cache and ring interconnect. On the far right is an integrated graphics processor which functions as a less powerful GPU; in the top left is the memory controller which sends data to and from the DRAM; and finally on the far left is the system agent and platform I/O, which communicates with the Chipset on the motherboard and manages the flow of data between many of the other components in your PC.

This motherboard is a massive, printed circuit board with thousands of wires running inside and a variety of microchips, components, sockets, ports, slots, headers, and connectors soldered to it. Perhaps the single most important and expensive component, aside from the bare PCB, is the Chipset, which is this integrated circuit found underneath a heat sink down here and is connected directly to the system agent section in the CPU. Here’s a diagram illustrating how the CPU and chipset are connected to everything else. As seen here, the CPU connects directly to the DRAM, one or two displays, the GPU, and perhaps a few SSDs plugged into the M2 slots. The Chipset manages most everything else: data flowing through the ethernet or Wi-Fi, data going to and from solid state drives and hard drives plugged into the SATA ports, some of the PCIe slots, your keyboard and mouse, USB devices, and the audio sent to the speakers or from the microphone.

Just a quick note, computer hardware has evolved immensely over the past 65 years and continues to evolve, so the details we show should be thought of as a current day example PC and not as how all computers work. Let’s move on and skip over the many different sockets and connectors throughout the motherboard and focus on the voltage regulator module or VRM found near the CPU. These components are used to drop the voltage coming from the power supply down to the 1.3 volts used by the CPU. As a result of all the power flowing through these components and their 80 to 90 percent efficiency, heat sinks need to be placed on top. While we’re on the topic of power, this CPU consumes power equivalent to approximately 16 LED bulbs, thus generating a considerable amount of heat which is taken away by the CPU cooler. This particular cooler uses a pump to circulate liquid through these tubes and into the radiator’s channels which transfers heat to the radiator fins. The fans then help transfer the heat to the air and the cooled liquid returns to the pump via the return loop. The pump is a brushless DC motor constructed from a PCB, a control chip, and a stator on the dry side, a barrier in the middle, and then the permanent magnet rotor

and impeller on the liquid side. There’s no mechanical connection between the rotor and stator, thereby preventing any leaks of the cooling liquid.

Let’s move onto the power supply which distributes power throughout the computer. In here, the main transformer reduces the voltage and bridges the isolation boundary between the primary side high voltage and the secondary side lower voltages used throughout your computer. Here’s the control PCB that ensures a stable output voltage and sends adjustment signals to the switching power transistor on the primary side using opto-isolators. There are dozens of other components used to filter the input voltage and generate various output voltages which are then sent to all the different hardware in your computer. For example, an SSD consumes just a few watts, and the connector uses these voltages, whereas your GPU can consume hundreds of watts using these connectors and these voltages.

Next, we’ll explore the GPU, but before that, let’s take a step back and consider the technology we’ve covered thus far. Alone, each of these components doesn’t do much at all, but when united, they combine to form a powerful system. Similarly, this video is a multidisciplinary combination of engineering, technology, art, and animation inspired by the Magic School Bus.

Let’s get back to dissecting this computer and move onto the graphics card and GPU which is also the brain of the computer… well, actually this analogy doesn’t really work very well but anyway… Opening the graphics card we see another PCB, with the GPU’s integrated circuit in the center, VRAM chips all around, and the voltage regulator module on the side. Above the IC is a heat sink with a fan to dissipate heat, and on the side of the graphics card are HDMI and Display Ports, a PCIe Interface, and then on the other side is the input power connector.

Let’s focus on the GPU integrated circuit which, similar to the CPU, has more than a thousand solder pads that connect it to the PCB. Opening the packaging we find the GPU die which is noticeably different from the CPU. In here are approximately 11.8 billion transistors organized into 6 graphics processing clusters, totaling 28 streaming multiprocessors. Each streaming multiprocessor is composed of 128 cores, resulting in a total of 3,584 cores. Each core has sections for integer and floating-point arithmetic, and sections for queueing in the operands and collecting the results and is far simpler than a CPU core. Additionally, on the die is an L2 memory cache shared among all the graphics processing clusters, a set of memory controllers that connect to the VRAM located around the processor, and a PCIe interface for connecting to the CPU.

When we zoom in to see a nanoscopic view of the integrated circuit we find something very similar to what we saw in the CPU, with the transistors on the bottom and a labyrinth of multiple layers of metal wires above. All these structures are manufactured in multibillion dollar semiconductor fabrication plants or fabs but that’s a topic for a whole different video, so let’s move on. GPUs and CPUs are similar in many ways, however, GPUs have thousands of cores that are limited to basic arithmetic, whereas CPUs have only a handful of cores that perform far more complicated operations. Additionally, CPUs have branch prediction and deep pipelines that optimize the execution of code. Here’s a quick example of what GPUs can do. Take this image comprised of 16 million pixels each with RGB values for each pixel. A simple way to brighten the image is to add 20 to each of these numbers. A CPU has 10 cores and thus performs arithmetic 10 numbers at a time whereas GPUs distribute the data to thousands of cores thus performing magnitudes more parallel processing

Let’s close this graphics card and talk briefly about the 3D model. Note that this is a slightly older model graphics card because we buy most of our hardware as non-functioning parts from eBay, and then, as shown earlier, we tear them down rather destructively in order to accurately model everything. Additionally, sometimes components shown aren’t compatible, such as this motherboard that says DDR4, while the DRAM is DDR5.

That said, modeling and animating everything to make it feel like you’re actually inside a computer took us around 500 hours and we would greatly appreciate it if you could take a few seconds to hit that like button, subscribe if you haven’t already, type up a quick comment below, and share this video with someone who will enjoy it. Also, we have a Patreon and would appreciate any support. We’re planning more videos on computer architecture and other related topics and can’t do it without your help, so thank you for doing these 4 quick things. It helps a ton.

Let’s move on and look at the DRAM, Solid State Drives, and Hard Drives. We’re not going to spend too long, because we have an entire series of videos on solid state drives, and then separate videos covering DRAM, and Hard Drives. But quickly, the CPU communicates directly with the DRAM through memory channels running inside the motherboard. Inside each of these 8 DRAM chips is an integrated circuit composed of 32 memory banks each 8192 columns wide by 65536 rows tall. The DRAM temporarily stores data using capacitors and transistors called 1T1C memory cells in 2D arrays that look like this. Data can be accessed within nanoseconds, however, among these 8 chips, only 16 Gigabytes of data can be temporarily stored. Take a look at our 35-minute video on DRAM, but for now let’s move on to SSDs which permanently store data in 3D arrays called 3D NAND. This array is 100 to 200 layers tall, 32 to 64 thousand columns wide and 32 to 64 thousand rows deep. Additionally, within a single SSD chip such as this one, are multiple 3D NAND arrays stacked one on top of the other. As a result, a single microchip can store terabytes of data, however reading or writing data takes 50 or so microseconds, which is 3000 times slower than DRAM. Zooming in on a single SSD memory cell we find a charge trap which stores different levels of charge allowing for 3 bits of data to be more permanently saved. Looking at the NVMe and SATA SSD, both have a few 3D NAND data storage chips, a DRAM chip for buffering and holding the data mapping table and a controller chip.

Let’s move on and dissect the hard disk drive. Here we have a disk mounted to a spindle with a motor that rotates the disk at thousands of rpm. A read write head moves across the disk in order to access a single track out of half a million other data tracks. Let’s zoom in on the read write head. The write head changes the direction of localized magnetic domains in a small layer on the disk whereas the read head senses these changes in magnetic domains. This disk drive is even slower than the SSD, taking a few milliseconds to access, thus resulting in slower read and write times, but costing less per terabyte of storage.

Thus far we’ve covered all the hardware inside your computer. So, thanks for watching this far, and as a bonus here’s what it looks like inside a computer mouse, with the scroll wheel up top, the infrared light, image sensor, and multiple lenses down here and the battery and processor in the middle. For computer mice we have separate dedicated videos exploring the image sensor and scroll wheel with incredible details. Additionally, here’s what it looks like inside a basic keyboard, with plastic traces that carry electricity to each key and, when pressed, that key completes a circuit which is sensed by the processor up here.