Höchstleistungsrechenzentrum Stuttgart: AMD Instinct™ GPU Training

Prerequisites and content levels

Prerequisites

Some knowledge in GPU programming, e.g., CUDA, and AI/Deep Learning

See also the suggested prereading below (resources and public videos).

Content levels

Basic: 1 hours
Intermediate: 4 hours
Advanced: 6 hours

Learn more about course curricula and content levels.

Resources

ROCm platform: with instructions for installing from Debian/CentOS/RHEL binary repositories. It also has links to source repositories for all components, including HIP.
HIP porting guide
ROCm/HIP libraries
ROC-profiler: Collects application traces and performance counters. The trace timeline can be visualized with chrome://tracing .
AMD GPU ISA docs and more

Public videos

Instructors

Essam Morsi, Adil Lashab and Philipp Samfass (AMD)

Agenda

All times are CEST.

Day 1

08:45 - 9:00 Drop in to Zoom

9:00 - 10:30 Introduction AMD GPU
- GCN/CDNA Overview
- Memory Hierarchy
- HIP GPU Compute Terminology
- Compute Units
10:45 - 12:45 HIP
- Introduction to HIP & Core HIP API
- Memory Management in HIP
- Asynchronous Computing with HIP
- Tips & Tricks
12:45 - 13:45 Lunch Break
13:45 - 15:15 ROCm
- Introduction to ROCm
- Multi-GPU RCCL/MPI with ROCm
- Debugging
- Profiling
15:30 - 16:15 AI
- RocOps introduction (AMD AI SW Stack)
- Training
  - Single GPU training with TF and PY for all models
  - mGPU training with TF and PY
  - Distributed mNode training with TF and PY
- Inference
  - MLPerf MIGraphX & TVM Backend

Day 2 (Hands-on)

08:45 - 9:00 Drop in to Zoom

9:00 - 9:45 ROCm setup
10:00 - 12:00 Hipification: CUDA To HIP
- Square
- RTM TTI
12:00 - 13:00 Lunch Break
13:00 - 14:30 HIP
- Profiling HW counters
- Example with bank conflicts on shared memory / how to profile / resolve it
- RocProfiler RocTracer ISA code
14:45 - 16:00 AI
- Distributed Training Demonstration
16:00 - 16:15 QA

Registration information

This course is already fully booked.

Fees

This course is free of charge.

Contact

Khatuna Kakhiani phone 0711 685 65796, kakhiani(at)hlrs.de
Lorenzo Zanon phone 0711 685 63824, zanon(at)hlrs.de

PRACE PATC and bwHPC

HLRS is part of the Gauss Centre for Supercomputing (GCS), which is one of the six PRACE Advanced Training Centres (PATCs) that started in Feb. 2012.

HLRS is also member of the Baden-Württemberg initiative bwHPC.

This course is also provided within the framework of the bwHPC training program. This course is not part of the PATC curriculum and is not sponsored by the PATC program.

Official course URL

http://www.hlrs.de/training/2022/GPU-AMD

Further courses

See the training overview and the Supercomputing Academy pages.