GPU Programming with CUDA and Python Training in Kraków, Poland

This course introduces participants to GPU programming concepts using NVIDIA CUDA and Python. Through hands-on examples, attendees will learn how to write parallel programs, offload computations to the GPU, and optimize performance using Numba, CuPy, and other CUDA-Python tools. The course is designed for Python developers who want to accelerate numerical computations and data processing using GPU resources.

INTRODUCTION TO GPU COMPUTING AND CUDA-PYTHON

Introduction to GPU Programming Concepts

  CPU vs GPU architecture

  What is CUDA and how it fits into Python workflows

  Use cases: image processing, scientific computing, ML acceleration

Python Tools for CUDA Programming

  Overview of Numba, CuPy, PyCUDA

  Installing and verifying CUDA environment

  Running basic GPU-accelerated code in Python

Getting Started with Numba CUDA

  Writing CUDA kernels using Numba

  Launch configuration: blocks, threads, grid

  Data transfer between CPU and GPU

GPU MEMORY, VECTORIZATION, AND ADVANCED KERNELS

Memory Management and Performance Optimization

  Types of GPU memory: global, shared, constant

  Optimizing memory access patterns

  Minimizing CPU-GPU data transfer overhead

Vectorization and CuPy for NumPy Users

  Migrating NumPy code to CuPy

  Universal functions (ufuncs) and element-wise operations

  Broadcasting, slicing, and array math on GPU

APPLICATIONS, DEBUGGING, AND PROFILING

Debugging and Profiling CUDA-Python Code

  Using cuda.profile and nvprof with Python

  Measuring kernel execution time and memory usage

  Common performance bottlenecks and solutions

Real-World Applications and Mini-Projects

  Case studies: matrix multiplication, convolution, or financial simulations

  Implementing and optimizing a sample GPU-accelerated project in Python

  Comparing CPU-only vs GPU-enhanced solutions

Final Wrap-Up and Next Steps

  Best practices for GPU Python workflows

  When to use GPU and when not to

  Resources to continue learning CUDA and parallel programming