LearnViz CompSci
Home / Learn / Computer Architecture & Logic / GPU Architecture

Advanced Topic 40 of 51 Computer Architecture & Logic

GPU Architecture

Massively parallel processors for throughput workloads.

How it works

A GPU contains thousands of simple cores grouped into streaming multiprocessors that run the same instruction across groups of threads called warps or wavefronts, following a single instruction, multiple thread model. It has high-bandwidth memory, many registers and fast shared memory per block, and hides latency by switching among many resident threads rather than by deep caches and speculation. Throughput is prioritized over single-thread speed.

Step-by-step walkthrough

  1. The CPU launches a kernel with thousands of threads.
  2. Threads are grouped into warps that run in lockstep.
  3. Thousands of simple cores hide memory latency by switching warps.
  4. Data comes from fast high-bandwidth memory.
  5. GPUs favor throughput over single-thread speed.

Where it is used

Graphics rendering, machine-learning training and inference, scientific simulation, video encoding and cryptocurrency mining use GPUs through CUDA, OpenCL or compute shaders.

Common pitfall

Branch divergence within a warp serializes paths, and uncoalesced memory access wastes bandwidth. Transfers between host and device memory over PCIe can dominate. Small or irregular tasks do not saturate the hardware. Design kernels with large parallelism, regular access and minimal data movement.

Where it fits in the course

GPU Architecture is topic 40 of 51 in Computer Architecture & Logic, rated Advanced. It follows SIMD and Vector Processing. The next topic is DMA.

Practice and explore

How to study this topic

  1. Read the explanation above and restate the definition in your own words.
  2. Open the interactive lesson and answer its mini quiz.
  3. Read the common pitfall, then write a short example that avoids it.
  4. Check your understanding in the Computer Architecture & Logic quiz, and revisit the topic if you miss a question.

Frequently asked questions

What is GPU Architecture?
Massively parallel processors for throughput workloads. It belongs to Computer Architecture & Logic and is an advanced-level topic.
Where do people use GPU Architecture?
Graphics rendering, machine-learning training and inference, scientific simulation, video encoding and cryptocurrency mining use GPUs through CUDA, OpenCL or compute shaders.
What is the most common mistake with GPU Architecture?
Branch divergence within a warp serializes paths, and uncoalesced memory access wastes bandwidth.
What should I learn before this topic?
Start with SIMD and Vector Processing, the topic just before it in Computer Architecture & Logic.
How can I practice it?
Answer the mini quiz in the interactive lesson, then take the Computer Architecture & Logic quiz.
Is GPU Architecture hard to learn?
It is an Advanced topic. Expect to read the explanation more than once and to connect it with earlier topics. Working through a small example by hand helps most.
How long does it take to learn GPU Architecture?
Most learners can read the explanation and finish the interactive lesson in under an hour. Fluency comes from applying the idea, so plan a few short practice sessions with the quiz.
Which topics are related to GPU Architecture?
SIMD and Vector Processing, DMA, Cache Coherence are closely connected. The full list is in the related topics section of this page.
What should I study after GPU Architecture?
Continue with DMA, the next topic in Computer Architecture & Logic.

Related topics