What is a GPU (Graphics Card) and How Does it Differ from a CPU?

Differences Between CPU and GPU Affecting Computer Performance

In modern computing, dominated by artificial intelligence (AI) and photorealistic graphics, two acronyms are often mentioned: CPU (Central Processing Unit) and GPU (Graphics Processing Unit). Although both are semiconductor chips performing mathematical operations, their engineering philosophies differ fundamentally. To understand why a high-end graphics card can cost over 2000 euros and why it is indispensable for training neural networks, it is essential to look at their microarchitecture. It’s not just a matter of speed – they represent different approaches to problem-solving.

Sequential Processing and Low Latency

The Central Processing Unit (CPU) is the system’s coordinator. Its architecture is focused on low latency and the ability to switch tasks quickly. Imagine the CPU as a highly skilled company manager or a mathematics professor: it can solve very complex, logically interconnected problems, but it does so sequentially – one after the other.
[Image of CPU vs GPU architecture diagram]
Modern consumer CPUs (e.g., Intel Core i9 or AMD Ryzen 9) typically have between 8 to 24 cores. Server-class processors (e.g., AMD EPYC) can have up to 96 or 128 cores. Each of these cores is powerful, has a large cache (L3 cache), and is designed to execute complex instructions, manage the operating system, and ensure smooth application launches.

Parallel Computing Flow

The Graphics Processing Unit (GPU) operates on a completely different principle. If the CPU is a manager, the GPU is an army of thousands of workers. Individually, a single GPU core is slower and simpler than a CPU core, but their quantity is staggering. The strength of the GPU lies in its high throughput.
For example, the high-end NVIDIA GeForce RTX 4090 graphics card has 16,384 CUDA cores. This allows it to perform thousands of calculations simultaneously. This architecture, known as SIMD (Single Instruction, Multiple Data), is ideally suited for tasks that are repetitive and can be broken down into smaller parts – such as calculating the colors of millions of screen pixels at once.

Comparison of Architecture and Purpose

To correctly choose equipment for work or entertainment, it is important to understand these technical nuances:

• Core Logic: The CPU uses a few but complex cores with sophisticated instruction prediction (branch prediction). The GPU uses thousands of simple cores designed for ‘green’ computing power.
• Memory Bandwidth: The GPU uses extremely fast memory (e.g., GDDR6X), with bandwidth reaching 1008 GB/s, while standard system RAM (DDR5) with the CPU typically achieves around 50–100 GB/s. This is essential for moving massive data arrays.
• Flexibility: The CPU is universal – it can run any code. The GPU is specialized – it is efficient only when the task can be divided for parallel processing.

The GPGPU Revolution and Matrix Computation

Until around 2006, GPUs were used almost exclusively for video games and graphic design. A breakthrough occurred with the advent of technologies like CUDA, which allowed programmers to access GPU cores not just for graphics but also for general-purpose computing (GPGPU).
It turned out that artificial intelligence, especially deep learning, relies on matrix multiplication – an operation that GPUs perform exponentially faster than CPUs. While a CPU might take weeks to train a complex neural network, a powerful GPU cluster can accomplish this in just a few hours. This is why today’s data centers are focused on GPU computing power, and companies like NVIDIA have become dominant in the tech market.