wunder beta

📘 How GPUs Work

A graphics processing unit is a processor built to apply similar operations to many data items at once. A scene may contain millions of vertices and pixels, and a scientific calculation may contain a large array of numbers. Instead of using

7
lessons
~30 min
to learn
Adults
level
Start the course →

What you’ll learn

  1. The GPU's jobExplain why GPUs use many parallel arithmetic resources for regular throughput workloads.GPUs trade general-purpose control complexity for high throughput across many similar data items.
  2. Inside the GPUIdentify compute groups, arithmetic lanes, schedulers, and resident thread state.A GPU repeats compute groups and keeps many thread groups ready to hide latency.
  3. Threads and warpsDescribe grids, blocks, warps, SIMT execution, and branch divergence.Threads are scheduled in related groups, so similar control flow keeps lanes productive.
  4. The graphics pipelineTrace vertices through rasterization, shaders, and framebuffer output.Graphics stages turn geometry into covered samples and calculated pixel values.
  5. General-purpose GPU computingExplain kernel launches, independent blocks, and shared-memory cooperation.GPU computing maps a repeated function over a grid while blocks cooperate locally.
  6. Memory and bandwidthRelate VRAM, caches, coalescing, locality, and bandwidth to GPU performance.Fast GPU arithmetic still depends on moving and reusing data efficiently.
  7. Specialization and performanceTrace a pixel through the pipeline and explain what determines real GPU speed.Graphics stages create parallel fragment work; specialized units and memory locality decide how much of the chip stays busy.

Questions this course answers

Why can branch divergence reduce GPU throughput?

A warp or wavefront shares an instruction stream, so different branch choices can require separate passes with inactive lanes.

Put these simplified graphics stages in order.

Geometry is transformed first, coverage creates fragments, shaders calculate their appearance, and output operations store the result.

What is a kernel in GPU computing?

A kernel is device code run by a grid of threads, with each thread commonly handling a different data element.

Match each GPU concept to its role.

These concepts live at different levels: device storage, execution grouping, block-local reuse, and programmable work.

Why is a GPU not automatically faster than a CPU for every program?

GPU throughput depends on how well the program maps to groups of similar threads, local reuse, bandwidth, and available specialized units.

Grounded in trusted sources

  • NVIDIA, CUDA Programming Guide, SIMT kernels and thread hierarchy: https://docs.nvidia.com/cuda/cuda-programming-guide/
  • NVIDIA, CUDA C Programming Guide archive, GPU architecture and memory hierarchy: https://docs.nvidia.com/cuda/archive/12.5.0/cuda-c-programming-guide/index.html
  • AMD GPUOpen, GPU architecture programming documentation: https://gpuopen.com/amd-gpu-architecture-programming-documentation/
  • AMD ROCm Programming Guide, programming model and memory: https://rocm-handbook.amd.com/_/downloads/amd-rocm-programming-guide/en/latest/pdf/
  • Microsoft Learn, Direct3D 12 graphics pipeline: https://learn.microsoft.com/en-us/windows/win32/direct3d12/
  • Wikimedia Commons MediaWiki API image records: https://commons.wikimedia.org/w/api.php

Every Wunder lesson is built from real, reputable sources — never invented.

Related courses

📘How RAM Works📘How SSDs Work📘How Wi-Fi Works📘How Fiber Optic Internet Works📘How DNS Works📘How HTTPS Works📘How Email Works📘How VPNs Work

Wunder is a personalized learn-anything platform — tell it any topic and it builds a beautiful, fact-checked course in minutes, with narration, a knowledge check, and a college-style University track.

All topics · Home

© 2026 Wunder Learning LLC · Terms & Privacy