HLRS High Performance Computing Center Stuttgart: Multi-GPU Deep Learning

Prerequisites and content levels

Prerequisites

For day one, you need basic experience with C/C++ or Fortran. Suggested resources to satisfy prerequisites: the learn-c.org interactive tutorial, https://www.learn-c.org/. Familiarity with MPI is a plus.

On day two, you need an understanding of fundamental programming concepts in Python 3, such as functions, loops, dictionaries, and arrays; familiarity with Pandas data structures; and an understanding of how to compute a regression line.
Suggested resources to satisfy prerequisites: Python Beginner’s Guide. Familiarity with TensorFlow and Keras will be a plus as it will be used in the hands-on sessions. For those who did not use these before, you can find tutorials here: github.com/tensorflow/docs/tree/master/site/en/r1/tutorials/keras.

Experience with Deep Learning using Python 3 and, in particular, gradient descent model training will be needed on day three and four. Further, expericen with PyTorch will be helpful, see https://pytorch.org/tutorials/ for instance.

Please be aware that while the second day offers an introduction or recap of Deep Learning most of the topics in this course are rather advanced. If you are completely unfamiliar with Deep Learning, the learning curve might be steep on days three and four.

Content levels

Basic: 4 hours
Intermediate: 4.5 hours
Advanced: 15 hours

Learn more about course curricula and content levels.

Instructors

Lecture and assistant trainers: PD Dr. Juan Durillo Barrionuevo (LRZ and NVIDIA University Ambassador), Tobias Haas, Khatuna Kakhiani, and Lorenzo Zanon (HLRS).

Learning outcomes

1^st day: Introduction to multi-GPU programming

You will have gained basic knowledge about how multi-GPU programming works.

2^nd day: Introduction to Deep Learning

Implement common deep learning workflows, such as image classification and object detection.
Experiment with data, training parameters, network structure, and other strategies to increase performance and capability.
Deploy your neural networks to start solving real-world problems.

3^rd day: Data Parallelism: How to Train Deep Learning Models on Multiple GPUs

Understand how data parallel deep learning training is performed using multiple GPUs.
Achieve maximum throughput when training, for the best use of multiple GPUs.
Distribute training to multiple GPUs using Pytorch Distributed Data Parallel.
Understand and utilize algorithmic considerations specific to multi-GPU training performance and accuracy.

4^th day: Model Parallelism: Building and Deploying Large Neural Networks

Train neural networks across multiple servers.
Use techniques such as activation checkpointing, gradient accumulation, and various forms of model parallelism to overcome the challenges associated with large-model memory footprint.
Capture and understand training performance characteristics to optimize model architecture.
Deploy very large multi-GPU models to production using NVIDIA Triton™ Inference Server.

Agenda

- preliminary -

1^st day (Tue): Introduction to multi-GPU programming (13:00 - 17:00)

On the first day you will learn the basics of multi-GPU programming. This will give you a rough idea how Deep Learning can be implemented using multi-GPUs.

2^nd day (Wed): Introduction to Deep Learning (9:00 - 17:00)

Explore the fundamentals of deep learning by training neural networks and using results to improve performance and capabilities.

During this day, you’ll learn the basics of deep learning by training and deploying neural networks.

3^rd day (Thu): Data Parallelism: How to Train Deep Learning Models on Multiple GPUs (9:00 - 17:00)

The computational requirements of deep neural networks used to enable AI applications like self-driving cars are enormous. A single training cycle can take weeks on a single GPU or even years for larger datasets like those used in self-driving car research. Using multiple GPUs for deep learning can significantly shorten the time required to train lots of data, making solving complex problems with deep learning feasible.

On the third day we will teach you how to use multiple GPUs to train neural networks.

4^th day (Fri): Model Parallelism: Building and Deploying Large Neural Networks (9:00 - 17:00)

Introduction to Training of Large Models,
Model Parallelism: Advanced Topics,
Inference of Large Models.

Exercises

The exercises will be carried out on cloud instances and on one of HLRS's clusters (on the first day).

HLRS concept for on-site courses

Besides the content of the training itself, another important aspect of this event is the scientific exchange among the participants. We try to facilitate such communication by

a social event on the evening of the first course day,
offering common coffee and lunch breaks and
working together in groups of two during the exercises.

Registration information

Register via the button at the top of this page.
We encourage you to register to the waiting list if the course is full. Places might become available.

Registration closes on June 16, 2024.

If you are not interested in all days, please select only those days in which you are interested while registering.

Important Information: After you are accepted, please create an account under courses.nvidia.com/join.

NVIDIA Deep Learning Institute:

The NVIDIA Deep Learning Institute delivers hands-on training for developers, data scientists, and engineers. The program is designed to help you get started with training, optimizing, and deploying neural networks to solve real-world problems across diverse industries such as self-driving cars, healthcare, online services, and robotics.

Fees

This course is open to academic participants only.

Students without master’s degree or equivalent: 0 EUR
PhD students or employees at a German university or public research institute: 0 EUR
PhD students or employees at a university or public research institute in an EU, EU-associated or PRACE country other than Germany: 0 EUR.
PhD students or employees at a university or public research institute outside of EU, EU-associated or PRACE countries: 0 EUR

Our course fees include coffee breaks (in classroom courses only).

For lists of EU and EU-associated countries, and PRACE countries have a look at the Horizon Europe and PRACE website.

Contact

Tobias Haas phone 0711 685 87223, tobias.haas(at)hlrs.de

HLRS Training Collaborations in HPC

HLRS is part of the Gauss Centre for Supercomputing (GCS), together with JSC in Jülich and LRZ in Garching near Munich. EuroCC@GCS is the German National Competence Centre (NCC) for High-Performance Computing. HLRS is also a member of the Baden-Württemberg initiative bwHPC.

This course is provided within the framework of the bwHPC training program.

Further courses

See the training overview and the Supercomputing Academy pages.

High-Performance Computing Center Stuttgart

Multi-GPU Deep Learning

Location