A GPU (Graphics Processing Unit), as the name suggests, was originally designed to draw images on screen. Calculating the colour of millions of pixels in every frame of a game requires the same simple operation to be performed on huge amounts of data at once. That parallel processing power has made GPUs indispensable in fields from scientific computing to artificial intelligence.
The difference between a CPU and a GPU
The central processing unit (CPU) is like the computer’s “general manager”: with powerful cores, usually numbering from a few to a few dozen, it runs the operating system and handles complex, sequential tasks that require decisions. A GPU is made up of thousands of simpler cores and excels at performing the same operation on lots of data simultaneously. As an analogy: a CPU is a few expert professors, while a GPU is a huge class of students each doing simple sums at the same time.
How GPUs got into AI
Training artificial neural networks is essentially multiplying enormous tables of numbers (matrices) — exactly the kind of parallel work GPUs do fast. In the mid-2000s, tools that made it easier to program GPUs for general computing (such as Nvidia’s CUDA platform) became widespread. The success of AlexNet, a model trained on GPUs, in an image recognition competition in 2012 is regarded as the start of the deep learning era. Today, large language models are developed through training runs lasting weeks or even months on thousands of interconnected GPUs in data centres.
Data centre GPUs and global competition
Data centre GPUs designed for AI, with high-bandwidth memory and fast chip-to-chip connections, are very different from — and far more expensive than — gaming cards. Demand for these chips has lifted chipmakers into the ranks of the world’s most valuable companies, and countries such as the US restrict exports of the most advanced AI chips on national security grounds.
Choosing a GPU for your own computer
- Integrated or dedicated? Integrated graphics built into the processor are enough for office work, browsing and video. Gaming, 3D design, video editing or running AI models locally require a dedicated graphics card.
- VRAM: The graphics card’s memory is decisive for high-resolution games and large AI models.
- Power and cooling: Powerful cards draw a lot of energy; make sure your power supply and case are suitable.
- NPUs: New laptops also include “neural processing units” (NPUs) that handle AI tasks with little energy; they are useful for everyday AI features.
Sources and further reading
- Nvidia, “What’s the Difference Between a CPU and a GPU?”.
- Krizhevsky, Sutskever and Hinton, “ImageNet Classification with Deep Convolutional Neural Networks” (2012).
- Intel, “CPU vs. GPU: What’s the Difference?”.
Leave a comment
Your email address will not be published. Required fields are marked *