A graphics processing unit (GPU) performs many similar calculations at the same time. This makes it especially effective for gaming, artificial intelligence and simulations, where large amounts of data can be processed independently and in parallel.
How a GPU works
A GPU contains many smaller processing units designed for parallel processing. Instead of completing a small number of complex tasks sequentially, it divides a large task into many similar operations and executes them simultaneously.
GPUs also provide high memory bandwidth, allowing large quantities of data to move quickly between GPU memory and processing units. This supports high-throughput workloads but does not mean that a GPU is always faster than a CPU.
| Application | Why a GPU is effective |
|---|---|
| Gaming | Pixels, vertices, textures, lighting and shaders require similar calculations across many parts of each frame. The GPU processes these calculations concurrently, increasing frame rate and visual detail. |
| Artificial intelligence | Training and running neural networks involve repeated vector and matrix operations. Thousands of these arithmetic operations can be distributed across GPU processing units. |
| Simulations | Weather cells, particles, fluids or physical objects can often be updated independently during each time step, making the calculations suitable for parallel execution. |
A central processing unit (CPU) has fewer, more powerful cores optimized for sequential processing, branching, operating-system tasks and general program control. A GPU has many simpler processing units optimized for high throughput.
The common misconception is that a GPU makes every program faster. Tasks with many dependencies, frequent conditional branches or little opportunity for parallelism may perform better on a CPU. Moving data between main memory and GPU memory can also create overhead.
IB exam technique
For A1.1 Computer hardware and operation, explain both the architecture and its consequence: many processing units enable simultaneous calculations, making GPUs suitable for data-parallel workloads. Avoid merely stating that GPUs are “more powerful”; identify parallelism, the type of repeated operation and a relevant application.
