What are GPU Cores?

What are GPU Cores?

A GPU, or Graphics Processing Unit, is a special type of processor designed to do many simple calculations at the same time. While a CPU has a few powerful cores that can handle complex tasks, a GPU has thousands of smaller, simpler cores that are great at doing the same calculation on lots of data simultaneously. This is why GPUs are so good at rendering graphics and training AI models.

GPU cores are organized into groups. In NVIDIA GPUs, these groups are called Streaming Multiprocessors, or SMs. In AMD GPUs, they are called Compute Units, or CUs. Each SM or CU contains a set of cores, along with shared memory and scheduling hardware. A modern GPU might have 80 or more SMs, each containing 128 cores, for a total of over 10,000 cores.

These cores are not like CPU cores. A CPU core is a complete processing unit that can handle any task independently. A GPU core is much simpler and works as part of a group. All the cores in an SM execute the same instruction at the same time, but on different pieces of data. This is called SIMD, Single Instruction Multiple Data, and it is the key to GPU efficiency.

Different types of cores handle different tasks. Shader cores handle the calculations for rendering 3D graphics, determining the color of each pixel. Tensor cores, found in NVIDIA RTX cards, are specialized for matrix math used in AI and deep learning. Ray tracing cores handle the complex calculations needed for realistic lighting and reflections. Each type of core is optimized for its specific job.

The number of cores is not the only thing that matters. Clock speed, memory bandwidth, and architecture efficiency all play a role. A GPU with fewer cores but higher clock speed and better architecture can outperform one with more cores but slower memory. This is why comparing core counts across different generations or different brands is not always meaningful.