Training Course: Fundamentals of Deep Learning & CUDA, May 2nd – May 4th 2022
Content
Accelerators are ubiquitous in today's HPC landscape, offering much higher performance compared to CPUs for large classes of applications. One of the predominant accelerator programming models used to develop parallel applications is CUDA by Nvidia. CUDA offers the basic functionality required for running applications on Nvidia GPUs, as well as making use of special hardware capabilities such as artificial intelligence support. Over the course of three days, participants will learn the basics of the CUDA programming model, how to make use of multiple GPUs for their applications, and how to use deep learning on Nvidia GPUs.
Every participant will use its own laptop/desktop to follow the hands-on sessions with a web browser connected to the Internet. The software and hardware required to complete the exercises and run the projects is provided via Amazon Web Services. Attendees will get free access to a fully configured GPU-accelerated server and have the opportunity to earn a certificate of subject matter competency issued by Nvidia.
The number of participants is limited to 40.
Details
Detailed Agenda and Content
Day 1: Fundamentals of Deep Learning ( detailed course description)
08:45 Registration
09:00 Welcome & Introduction
09:15 The Mechanics of Deep Learning
11:15 Break
12:15 Pre-trained Models and Recurrent Networks
14:15 Break
14:30 Transfer Learning
15:30 Advanced Architectures and Natural Language Processing
16:30 Final Review & Closing
17:00 End of day 1
Day 2: Fundamentals of Accelerated Computing with CUDA C/C++ ( detailed course description)
09:00 Welcome & Introduction
09:15 Accelerating Applications with CUDA C/C++
11:15 Break
12:15 Managing Accelerated Application Memory with CUDA C/C++
14:15 Break
14:30 Asynchronous Streaming and Visual Profiling for Accelerated Applications with CUDA C/C++
16:30 Final Review & Closing
17:00 End of day 2
Day 3: Accelerating CUDA C++ Applications with Multiple GPUs ( detailed course description)
09:00 Welcome, Introduction, and Application Overview
09:30 CUDA Streams for Copy/Compute Overlap
10:15 Multiple GPUs with CUDA C++
11:00 Break
11:15 Copy/Compute Overlap with Multiple GPUs
12:00 Course Assessment and Final Review
12:30 End of day 3
Requirements:
Basic programming skills in C/C++ are required, as well as an understanding of fundamental programming concepts in Python such as functions, loops, dictionaries and arrays. No previous knowledge of CUDA programming is assumed.
Lecturer:
Manuel Ujaldón, Nvidia DLI Instructor @ University of Malaga
Manuel was awarded CUDA Fellow by Nvidia in 2012, and since then, he has been the instructor of more than 130 activities related to HPC and Deep Learning in more than 20 countries, including more than 70 invited talks and tutorials in ACM/IEEE conferences.
In 2019, Manuel was recognized as DLI University Ambassador by Nvidia and currently organizes DLI courses worldwide. He will visit UIBK next May to teach as instructor three of the most popular full-day workshops of the DLI catalog: "Fundamentals of Deep Learning", "Fundamentals of Accelerated Computing with CUDA C/C++” and “Accelerating CUDA C++ Applications with Multiple GPUs”.
Language:
English
Time and Location:
Monday, May 2nd – Wednesday, May 4th 2022, 09:00 – 17:00
HSB 5 (ground floor of the Civil Engineering Building, "Hörsaaltrakt Bauingenieursgebäude")
Technikerstr. 13b, Technik Campus (map)
6020 Innsbruck
Austria
Registration:
TODO
Fee:
The course is free of charge
Course Material:
After enrolling, participants will have the opportunity to access and download the course material.
Contact Information:
Department of Computer Science
Philipp Gschwandtner, MSc PhD
Technikerstr. 21a
6020 Innsbruck
Tel.: +43 512 507-53233
philipp.gschwandtner@uibk.ac.at