CUDA cores are specialised processors built into Nvidia graphics cards (GPUs). Each core performs calculations in parallel, allowing the card to process thousands of small tasks simultaneously instead of one after another.
Nvidia designed CUDA cores specifically for parallel processing. They're smaller and less versatile than a CPU's cores, but when software is written to use them, they excel at graphics rendering, video encoding, machine learning, and scientific simulations. This parallel capability is what makes GPUs useful for tasks beyond gaming.
The number of CUDA cores varies widely across Nvidia's product range. A basic entry-level card might have 384 cores, whilst high-end professional models can have 10,000 or more. More cores correlate with higher throughput, but raw core count isn't the whole story. Clock speed, memory bandwidth, and the generation of the architecture also affect real-world performance.
When comparing graphics cards for video editing, 3D rendering, or AI work, check the CUDA core count alongside the memory capacity and power requirements. A card with more cores but slower memory might underperform against a newer, slightly less densely packed model. Similarly, if your software doesn't support CUDA acceleration, core count becomes irrelevant. Nvidia's official specifications always list CUDA cores, so you can compare models directly before buying.