LearnViz CompSci
Home / Learn / Concurrency & Parallelism / CUDA Programming

Advanced Topic 37 of 51 Concurrency & Parallelism

CUDA Programming

NVIDIA's model for general-purpose GPU computing.

How it works

CUDA is NVIDIA's platform for general-purpose GPU programming. The programmer writes kernels in a C++ extension that run across a grid of thread blocks, each containing many threads organized in warps of 32 that execute in lockstep, with indexes such as threadIdx and blockIdx identifying work items. The host manages device memory, copies data between CPU and GPU and launches kernels, with streams enabling overlap of transfers and computation. Libraries such as cuBLAS, cuDNN and Thrust provide optimized building blocks.

Where it is used

Deep learning training and inference, scientific simulation, image and video processing, cryptography, finance and graphics ray tracing use CUDA.

Common pitfall

Performance depends on coalesced global memory access, occupancy, use of shared memory, avoiding warp divergence and minimizing host-device transfers over PCIe. Race conditions and out-of-bounds accesses in kernels are hard to debug. Kernel launch overhead hurts tiny workloads. Profile with Nsight, use libraries where possible and handle errors from every call.

Where it fits in the course

CUDA Programming is topic 37 of 51 in Concurrency & Parallelism, rated Advanced. It follows MPI. The next topic is GPU Memory Hierarchy.

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 Concurrency & Parallelism quiz, and revisit the topic if you miss a question.

Frequently asked questions

What is CUDA Programming?
NVIDIA's model for general-purpose GPU computing. It belongs to Concurrency & Parallelism and is an advanced-level topic.
Where do people use CUDA Programming?
Deep learning training and inference, scientific simulation, image and video processing, cryptography, finance and graphics ray tracing use CUDA.
What is the most common mistake with CUDA Programming?
Performance depends on coalesced global memory access, occupancy, use of shared memory, avoiding warp divergence and minimizing host-device transfers over PCIe.
What should I learn before this topic?
Start with MPI, the topic just before it in Concurrency & Parallelism.
How can I practice it?
Answer the mini quiz in the interactive lesson, then take the Concurrency & Parallelism quiz.
Is CUDA Programming 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 CUDA Programming?
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 CUDA Programming?
MPI, GPU Memory Hierarchy, OpenMP are closely connected. The full list is in the related topics section of this page.
What should I study after CUDA Programming?
Continue with GPU Memory Hierarchy, the next topic in Concurrency & Parallelism.

Related topics