The Cybermedia Center, Osaka University will start the computing service by OCTOPUS, a hybrid-typed cluster system with 1.46 PFlops peak performance, composed of general CPU nodes, many-core nodes, GPU computing nodes, large-scale main memory nodes, and large-scale storage.
In this challenge, as a part of celebration of OCTOPUS introduction, we call for programs (non-commercial) that can be accelerated on the coming GPU computing nodes of OCTOPUS, so that GPU computing nodes are leveraged right after system introduction. The programs accepted in this challenge will be performance-tuned and optimized by the Cybermedia Center and NVIDIA experts.
GPU computing nodes of OCTOPUS has four NVIDIA GPU Tesla P100 equipped on NVLink high-speed interconnect network. This architecture provides a unified memory space of 64GB GPU memory(16GBx4) and CPU memory (192GB) for high parallelism.
This challenge aims to compare vector processing on SX-ACE and computing performance of Tesla P100. For the reason, this challenge targets programs currently executed on a single node of SX-ACE.
Please apply if you have any concern and interest to this challenge.
Please see the following page for the detail.
OCTOPUS Celebration GPU Challenge
Posted : October 03,2017