Frontier COE Workshop – Virtual
September 3, 2020
10:30 AM – 5:00 PM (ET)

This workshop is by strict invitation only.

The Oak Ridge Leadership Computing Facility (OLCF) will host a virtual Frontier COE Workshop on September 3, 2020. During the workshop, members of the Frontier Center of Excellence (COE) will give updates on the latest hardware/software on the Poplar and Tulip systems. For more details, please see the “Agenda” dropdown below.

Zoom Details – ECP’s FedRamp Zoom will be used to facilitate this event, and all participants will need to be manually admitted from the waiting room into the meeting due to NDA restrictions (see below).

We will be using the list of registered participants to admit people from the waiting room into the workshop, so YOU MUST REGISTER to attend the event (see “Registration” dropdown below). The deadline for registration is August 24 for non-US citizens and August 28 for US citizens..

The zoom details will be emailed to registered participants prior to the event.

Non-Disclosure Agreement (NDA) Restrictions – All sessions will contain and allow discussion of AMD and HPE proprietary information, so all attendees must be covered by an institutional NDA.

If you have any questions, please contact Tom Papatheodore (


Time (all times EDT) Topic Presenter
10:30 AM – 11:00 AM Zoom Admission
11:00 AM – 11:15 AM Welcome & Logistics Tom Papatheodore (OLCF)
11:15 AM – 11:30 AM Training Agenda
COE (re)Introduction
NDA Procedures Refresh
Poplar and Tulip Updates
Noah Reddell (HPE)
11:30 AM – 12:15 PM hip-Clang / ROCclr with ROCm 3.5 and beyond Nick Curtis (AMD)
12:15 PM – 1:00 PM AMD GPU Hardware Nick Malaya (AMD)
Rene van Oostrum (AMD)
1:00 PM – 1:10 PM Break
1:10 PM – 1:40 PM Achieving high STREAM efficiency on AMD GPUs Nick Curtis (AMD)
1:40 PM – 3:25 PM Cray PE for AMD GPU: beta capabilities for profiling and debugging Heide Poxon (HPE)
Andrew Gontarek (HPE)
3:25 PM – 3:35 PM Break
3:35 PM – 4:15 PM Lessons learned from CAAR and ECP teams PIConGPU, Kokkos, HACC
4:15 PM – 5:00 PM Comparison of GPU programming model implementations for nearest-neighbor exchange Trey White (HPE)