AMD EPYC Advanced User Training on Expanse

The complexity of the AMD EPYC architecture, with its large core counts, non-uniform memory access and distributed L3 caches, can make it challenging to obtain peak performance. We’ll cover a range of intermediate-to-advanced topics that will help you to make most effective use of Expanse and other EPYC-based systems. These include an overview of the EPYC architecture, AMD’s compilers and math libraries, strategies for mapping processes and tasks to compute cores, Slurm, application tuning and profiling tools.

To get the most out of this training event, you should already be familiar with the basics of working in a Linux environment, job submission, compilation and programming in C/C++, Fortran or other languages.

The event is open to all users with an XSEDE portal account. Given the amount of material and potentially large number of participants, we will not be doing hands-on exercises during the event. Attendees who have access to Expanse will be able to download exercises so that they can practice what they learned.

Schedule (Pacific time listed)

• 9:00 AM – 9:05 AM – Welcome & Introduction
  • Mary Thomas, Computational Data Scientist, SDSC
    
• 9:05 AM – 9:35 AM – Introduction to 2nd Gen AMD EPYC™ Processors
• 9:35 AM – 9:45 AM – Q&A session
  • Mark Klonower, Field Application Engineer, AMD
• 9:45 AM – 10:25 AM – AOCC Compiler & AOCL Math Libraries
• 10:25 AM – 10:35 AM – Q&A session
  • Marty Kandes, Computational and Data Science Research Specialist, SDSC
    
• 10:35 AM – 11:15 AM – SLURM, Runtime Configurations, and Tuning
• 11:15 AM – 11:25 AM – Q&A session
  • Mahidhar Tatineni, Director of User Services, SDSC
    
• 11:25 AM – 11:55 PM – 30-minute break 

• 11:55 AM – 12:35 PM – Profiling applications using AMD uProf
• 12:35 PM – 12:45 PM – Q&A session
  • Bob Sinkovits, Director for Scientific Computing Applications, SDSC
    
• 12:45 PM – 1:00 PM – Final Q&A and Wrap-up
  • Mary Thomas, Computational Data Scientist, SDSC