AnsweredAssumed Answered

Student Cluster Competition

Question asked by hofm on Feb 3, 2017
Latest reply on Feb 3, 2017 by goodplay



maybe this is the wrong forum to ask, so feel free to move this thread or let me know whom best to ask.


I'm one of the researches supporting one of this year's student teams participating at the student cluster competition in Frankfurt at ISC17. I also supported a team last year at the student cluster competition at SC16 in Salt Lake City, where we learned that one of the other teams was using AMD GPUs. The goal of the competition is to reach the best performance for a set of well-known applications, such as HPL, HPCG, etc. within a power constraint of 3000W.


HPE will be our main sponsor and supply us (probably) with several dual-socket Skylake-EP nodes; were currently in the process of evaluating different accelerators. Apart from HPL (which requires high DP flop/s that most consumer GPUs don't thrive at), there is HPCG (which is memory-bound and benefits from HBM/HBM2 technology) and PaddlePaddle (PArallel Distributed Deep LEarning), which benefits from FP16-capable architectures. I was wondering, whether AMD might be interested in offering some loaners for evaluation. The venue is of high visibility within the HPC community, as is the student cluster competition.