In collaboration with our partner AMD, an online introductory training on optimising AMD’s cutting-edge MI300 series GPUs was held. More than fourty participants were granted access to AMD resources and attended these sessions, delivered by AMD experts, where they gained valuable skills in: 

• New concepts for enhanced memory management to reduce data transfer times. 

• Techniques to unlock the full potential of these powerful GPUs and optimise research workflows. 

• The latest innovations from AMD that can accelerate research and development efforts. 

Session 1 

• Introduction to the System for Exercises 
• APU Programming Model 
• Introduction to OpenMP® Offloading 
• Real-World OpenMP® Language Constructs 
• Advanced OpenMP® – debugging and optimisation 

Session 2 

• HIP and ROCm 
• Porting code to HIP and Portable Build System 
• Optimising HIP Code 
• OpenMP® and HIP Interoperability 
• Intro to ML/AI Exercises 

Session 3 

• Timeline profiling with Rocprof and Rocprof-sys 
• Kernel profiling with Rocprof and Rocprof-compute 
• Overview of Other Tools – AMD and HPC Community 
• AI Programming Assistant 
• Additional Training Resources 

Any questions, please do not hesitate to contact: Richard Regan, DiRAC Training Manager