GPU (Graphics Processing Unit)

A GPU (graphics processing unit) is a specialized processor with thousands of small cores designed to perform many calculations in parallel. Originally built to render graphics, GPUs are now among the most widely used accelerator cards in high-performance computing (HPC) and are the dominant hardware for AI training and inference, because those workloads depend on large numbers of simple calculations running at the same time.

What is a GPU?

A GPU packs a very large number of comparatively simple cores onto a single device, potentially numbering in the thousands. This contrasts with a CPU, which has far fewer but far more sophisticated cores. GPUs were originally intended to expedite 3D graphics rendering, but their programmability grew over time, letting developers harness them for general-purpose computation well beyond graphics. Today the same hardware drives video games, video editing and content creation, scientific simulation, and deep learning.

GPU vs. CPU

GPU and CPU architectures share the same fundamentals but optimize for opposite goals. A CPU targets task parallelism, executing a modest number of independent, complex operations quickly, and devotes significant silicon to pipelining and other sequential-execution features. A GPU targets data parallelism, applying the same instruction across many data items at once (SIMD).

	GPU	CPU
Core count	Thousands of simple cores	8 to 128+ complex cores
Optimized for	Data parallelism (SIMD)	Task parallelism, sequential work
Best at	Many simple calculations at once	Complex, branching logic
Typical role	Accelerator	General-purpose processor

Why GPUs are used for HPC and AI

GPUs suit HPC because of their highly parallel architecture, specialized processing capability, and high memory bandwidth. Applications such as scientific simulations, machine learning, and data analytics require enormous compute, and a GPU performs the underlying arithmetic much faster and more energy-efficiently than a CPU can.

The same properties make GPUs ideal for AI. Training and inference rely on a very large number of simple calculations that can run in parallel, and much of a CPU's complexity is wasted on that work. A GPU's thousands of simple cores execute these calculations far more efficiently than a CPU's lower core count. The largest AI models pool many GPUs together to supply enough memory and compute for a single job.

How GPUs work with CPUs

In practice CPUs and GPUs work together. The CPU handles control flow, coordination, and sequential logic, then offloads the heavy parallel math to one or more GPUs. As AI demand grows, specialized machine learning accelerators and FPGAs are emerging to complement or move beyond general-purpose GPUs.