Literary

Cet atelier sera donné en anglais par Mr. Nikolai Sergueev, analyste en calcul scientifique à Calcul Québec à l'Université de Montréal. Summary This workshop tutorial introduces the Graphics Processing Unit (GPU) as a massively parallel computing device, the CUDA parallel programming language, and some of the CUDA numerical libraries for use in high performance computing. We start by demonstrating the difference between the Central Processing Unit (CPU) and GPU, then explain CUDA GPU architecture along with CUDA execution and memory model. Second part of the tutorial is a hands-on training session where we show how to compile and run CUDA C/C++ programs as well as use CUDA FFT and linear algebra libraries for intensive numerical computations. Upon completion, participants should understand key GPU concepts and be able to write simple GPU programs with CUDA C/C++. Prerequisite Knowledge of the C/C++ programming language is required. Attendees should bring their personal laptop computer to participate in the hands-on content. Lesson plan Understanding the architecture of a GPU Understanding the workflow of a CUDA program Understanding the core principles to use a GPU efficiently Identify whether an algorithm is a good candidate for ‘easy’ performance gains on a GPU Find information on the web and within the NVidia CUDA SDK Writing and compiling a minimal CUDA code and compiling CUDA examples Managing GPU memory and understanding the various types of GPU memory Using threads and blocks to write parallel algorithms Checking errors Overview of what’s outside CUDA (OpenACC, Libraries, OpenCL, Applications) Optional topics: Using multiple GPUs Using unified memory in Cuda 6 Nvidia Visual Profiler Asynchronous operations with Cuda streams FAQ Comment puis-je contacter l'organisateur si j'ai des questions? Vous pouvez contacter l'adresse support@calculquebec.ca