Learn about the intricate process of how a graphics card is made, from design to final quality checks. Discover the steps involved in creating this essential computer component.
How a Graphics Card is Made
As we did with processors, in this other article we are going to see how a graphics card is made , both the GPU and the PCB that composes it. This way you will be able to learn more about these elements that are so crucial for today’s computers, and that are so complex and beyond the understanding of some users who are new to this.
Index of contents
- PCB and GPU design
- GPU
- PCB Layout
- PCB Manufacturing
- GPU chip manufacturing
You may also be interested in knowing which are the best graphics cards on the market
PCB and GPU design
To manufacture a graphics card, the first thing is to design the GPU, which will be handled by the fabless like AMD, NVIDIA, etc., or the IDM like Intel, and on the other hand, there will be the PCB manufacturers, who will also design their own circuit board, which is nothing more than a “mini” motherboard, but instead of a CPU it uses a GPU, and instead of RAM it uses VRAM, as well as a graphic BIOS instead of the classic BIOS/UEFI…
Although I have left it out, obviously, other elements will also have to be designed and manufactured, such as the VRAM chips by companies like Micron, Samsung, SK Hynix, etc., other controller chips that are necessary for the operation of the card and that can be created by very different companies, as well as the electronic elements (capacitors, resistors, …).
GPU
To design the GPU , you must start by describing the performance, consumption, etc. objectives to be achieved, and based on these, begin to design the architecture. In this case, unlike the CPU, it is a massively parallel architecture, with hundreds or thousands of very simple cores called shaders. Each of these cores will contain texture units, rasterization units, and ALUs or FPUs, among others.
However, aside from these differences, the design is exactly the same as the CPU . In fact, although it is not often talked about, the GPU also has its own ISA or instruction set, but it is specially designed for graphics processing, and it works thanks to the drivers and the graphics APIs such as OpenGL, Vulkan, DirectX, or the OpenCL, CUDA APIs, etc. if working in GPGPU mode, translating the necessary calculations to be generated into a series of commands or instructions that the GPU will process. Therefore, if the designer sticks to supporting these APIs, they will work correctly with any system, regardless of whether the ISA is different.
For example, if a software makes a request to draw a graphic on the screen, that request will go to the graphics API, which will translate that request into a language appropriate for the GPU to understand. Once the GPU has processed it, the API returns the response to the software that requested it. In other words, thanks to the graphics API, the complexity of the GPU hardware is abstracted, allowing different GPUs with very different ISA and architecture to work well in the same system.
Of course, in order for this ISA to be processed, the microarchitecture must also be designed , that is, the physical implementation capable of processing the instructions and data defined by the ISA in such a way that it does so as quickly and efficiently as possible.
After that, an EDA environment and HDL languages will be used , just as it happened with the CPU, to describe the microarchitecture and circuit, in addition to performing tests through simulation, verification, etc., and finishing by translating the HDL code into basic cells, routing them and obtaining the final layout with which the masks for the manufacture of the chip in the foundry can be created…
PCB Layout
On the other hand, companies like MSI, Gigabyte, ASRock, ASUS, Sapphire, Zotac, PNY, PowerColor, EVGA, Biostar, etc., will be in charge of creating their own PCBs , which can be different. However, all of them must be compatible with the GPU chip, of course. To do this, AMD, NVIDIA, Intel, etc., will send a series of documentation so that you can know the contacts of the graphic chip and its functions (data, clock, voltage, GND, …).
When designing a PCB or printed circuit board, the first thing to do is to keep in mind what needs to be created, to make a list of the necessary electronic components (chips, capacitors, resistors, ports…), and thus determine how they are all interconnected so that they work properly.
Schematic design software tools such as KiCAD, Eagle, Altium Designer, etc. are used for the layout , although some of the companies mentioned may even have solutions developed by themselves. Either way, the schematic will include a series of symbols and connection lines that represent the circuit. In other words, the software allows you to place components on the board, define the copper traces that connect the components, create additional copper layers for power and ground signals, and add holes for mounting the board.
Once the schematic has been created, schematic capture is performed to link it to the PCB layout. This involves associating each symbol on the schematic with its corresponding physical component on the PCB. In addition, a number of physical rules or constraints will be defined for the layout, such as track widths, distance between tracks, hole sizes, space that can be occupied, etc.
Design checks must be performed to detect errors in the PCB layout, such as short circuits, open traces, and violations of design rules. Design verification software (DRC) tools are used to perform these checks. Finally, the layout can be passed to a series of templates that will be used for manufacturing the board…