Graphics Card
Overview
A graphics card is expansion hardware in a computer dedicated to computing images and video and outputting them to a display; its formal names are video card or display adapter. The GPU (Graphics Processing Unit), the heart of the card, arranges thousands of small cores in parallel to process pixel-level computations simultaneously, and thanks to this structural characteristic, since the 2010s it has established itself not only in 3D rendering but also as a core device for general-purpose parallel computing (GPGPU) such as AI training and inference, scientific simulation, and cryptocurrency mining. Today, the graphics card market is divided into the consumer gaming market and the data center/AI market, making it one of the biggest battlegrounds in the semiconductor industry.
Main Content
1. Hardware Components
A graphics card consists largely of a GPU chip (die), VRAM (video memory), power delivery (VRM), cooling solution, PCB substrate, and output interface. The GPU die is divided into compute cores (CUDA cores/stream processors), texture mapping units (TMU), raster operations units (ROP), ray tracing accelerators, tensor cores, and so on. VRAM is dominated by the GDDR family, with a generational shift from GDDR6 to GDDR7 underway, and HBM (high bandwidth memory) is used in top-tier products. Memory bus width and bandwidth are decisive metrics determining performance in processing high-resolution, high-quality textures. Power is supplied through the PCIe slot and auxiliary power connectors (8-pin, 12V-2x6, etc.), and the power consumption of recent flagship cards exceeds 400W.
2. History and Generations
In the early 1990s, 2D-only cards and 3D accelerator cards — represented by 3dfx Voodoo — existed separately, but in 1999 NVIDIA's GeForce 256 established the concept of the "GPU," opening the era of integrated 3D graphics cards. Since then it has developed through competition with ATI (now AMD), the emergence of the CUDA platform in 2006, the introduction of real-time ray tracing (RTX) in 2018, and the popularization of AI upscaling technologies such as DLSS and FSR in the 2020s. The interface settled on PCI Express (currently PCIe 5.0/6.0) after AGP, and HDMI and DisplayPort are the standard output ports.
3. Classification
By use, they are divided into consumer cards for gaming and creators, professional workstation cards (Quadro, Radeon Pro), data center/AI accelerator cards (Instinct, H100/B200 series), and integrated graphics (integrated GPU). In terms of form factor, there are desktop expansion cards, laptop MXM/onboard designs, and external GPU (eGPU) dock designs. Performance tiers are generally divided into entry (tier 60 and below), mid-range (tier 70), and high-end (tiers 80–90).
4. Major Manufacturers and Ecosystem
GPU design is led by three companies — NVIDIA, AMD, and Intel — with NVIDIA holding about 80–90% of the consumer market. Finished cards are manufactured by add-in board (AIB) partners such as ASUS, MSI, Gigabyte, Emtek, Sapphire, and PowerColor. The software ecosystem is also a key competitive factor: NVIDIA's CUDA has become the de facto standard for AI and scientific computing, with AMD's ROCm and Intel's oneAPI in pursuit. Graphics APIs include DirectX 12 Ultimate, Vulkan, OpenGL, and Metal.
5. Performance Metrics
Performance is evaluated by core count, clock speed, memory bandwidth, TDP (power consumption), and actual in-game frames and benchmarks. Recently, ray tracing processing performance, upscaling quality, and support for frame generation technology have been added as key variables in purchasing decisions. Performance per watt (frames per watt) and heat and noise are also important real-world metrics.
6. Applications
The range of uses is very broad: 3D games, video editing and 3D rendering, CAD, medical image processing, weather and molecular simulation, autonomous driving training, training and inference of generative AI models, and cryptocurrency mining. In particular, since 2022 the generative AI boom has caused demand for data center GPUs to explode, reshaping the industry structure.
Latest Trends
The biggest topics in the 2024–2025 graphics card market are, unsurprisingly, AI acceleration and power efficiency. NVIDIA launched the Blackwell architecture-based RTX 50 series (2025) and the data center B200/GB200, fully introducing GDDR7 memory, PCIe 5.0, and the 12V-2x6 power standard. The trend of AI being integrated throughout the rendering pipeline — such as DLSS 4's multi-frame generation — is accelerating; AMD is responding with the RDNA 4 architecture and FSR 4 (machine learning-based), while Intel is targeting the value market with the Arc B series (Battlemage). Meanwhile, continued US export restrictions on advanced GPUs to China have brought about regional differentiation in product specifications and a restructuring of supply chains, and the HBM supply competition between Samsung Electronics and SK Hynix, dependence on TSMC foundry, and surging demand for chiplet design and advanced packaging (CoWoS) have also emerged as major issues. In the consumer market, along with the spread of 4K and high refresh rate monitors, concerns about power consumption and rising prices are growing, and undervolting and efficiency-focused tuning have become a new user trend.
Related Topics
- [[GPU]]
- [[엔비디아]]
- [[AMD]]
- [[인공지능]]
- [[반도체]]
- [[CUDA]]