SURF likes to be close to our members and keeps abreast of new technologies that they use for research.

Workshop for Introducing AMD GPU Architectures

We are currently collaborating with University of Amsterdam, AMD and NIKHEF to organise a workshop session (online and in-person) for AMD GPUs. This workshop aims to cultivate the interest in the new AMD GPU architecture and it's use.

In this workshop and webinar sessions people will get an overview of the new AMD GPU architecture. We will discuss its performance, usability and programmability and with the assisted demos and hands-on exercises we provide you with the knowledge and tools to start working with your own code on AMD GPUs

For Whom

This workshop is mainly targeted at application developers, HPC support scientists and programmers wanting to get a glimpse of new AMD GPU architectures and its relevance for scientific computing.

Prior Knowledge required

• Knowledge of programming langauges like C/C++/Python or FORTRAN
• Basic experience with programming GPU devices
• Basic experience with working with Linux based server machines

Webinar Content (2 hours)

• Introduction to AMD GPU Architectures
• introduction to AMD GPU Porting Tool and Techniques
• Live porting demo
• Questions & Answers

Workshop Content (8 Hours)

• Introduction to AMD GPU arhictecture and software ecosystem
• Hands-on performance analysis for GPU node
• Experience of porting exercise for the next generation GPU architectures
• Programming models and Tools for porting applications on AMD GPUs
• Proof of concept applications
• Analysis and porting of user selected applications

Currently we have following nodes hosting the AMD GPUs

• 3 GPU Machine at NIKHEF

  • Hostname : viltstift.nikhef.nl
  • 2x vega 20 (MI50)
  • ROCM 3.0 and 3.1 with HCC backend

• AMD Vega 20 Machine

  • Hostname : sne-dtn-04.lab.uvalight.net
  • 1x vega 20 (Radeon VII)
  • ROCM3.0 with clang-11 backend

• 8 GPU Machine at NIKHEF

  • Hostname : kleurplaat.nikhef.nl
  • 8x vega 20 (MI50)
  • ROCM 3.0 with HCC backend
  • GCC 9.2.0
  • clang 9.0.1
  • OpenMPI 4.0