Follow along with self-paced training videos to learn about using the ROCm platform. 

Introductory ROCm™ Software Platform Training

Introduction to ML Frameworks

Introduction to AMD CDNA™ and AMD CDNA™ 2 architectures, plus demos of TensorFlow and PyTorch. 

Introduction to Omni Tools

Topics covered in this presentation include rocprof, Omnitrace, and Omniperf.

Introduction to Rocprof

Overview of ROC-profiler (rocprof) for AMD GPU profiling libraries, central components  for collection of application tracing and counter collection.

Introduction to the Architecture

This presentation goes over the AMD Instinct™ architecture and the basics of developing applications within the AMD ROCm ecosystem.

Application Training

Developing Fortran Applications

Learn about Hipify -Fortran with separate CUDA® routines, HIPFort, and using OpenMP off loading a directive-based approach, OpenACC alternative.

Getting Started with OpenMP®

Learn about the AMD open software stack and building simple OpenMP offload applications on AMD Instinct accelerators. 

Porting Applications to HIP

Learn how to use Hipify-perf and Hipify-clang to port applications to HIP.

Advanced ROCm Training

AFFINITY – Placement, Order, and Binding

This training looks at modern Heterogenous architectures, what is AFFINITY and why it is important, and understanding of node topology.

AMD Debugger: ROCgdb

Learn about the source level debugger in ROCm and GNU debugger (GDB).

AMD Node Memory Model

Learn about coarse/fine grain memory, floating point (FP) hardware atomics in HIP, and view a preliminary performance study of course vs fine grain memory.

GPU Aware MPI with ROCm

This presentation discusses Running GPU-ware MPI examples on the LUMI cloud and measuring GPU-Aware communication bandwidth and latency.

Hierarchical Roofline on AMD Instinct MI200 GPUs

This presentation will cover Roofline fundamentals, empirical hierarchical Roofline on MI200 and Omniperf with examples.

OpenMP Offload Programming

This presentation covers avoiding dates transfers, asynchronous offloading, and hybrid programming with OpenMP and HIP.

Code Labs

Omniperf Code Lab #1
Omniperf Code Lab #2
Omnitrace Code Lab