Ouessant: Introduction

The first phase of the OpenPOWER prototype, Ouessant, arrived at the beginning of February, 2016. It consisted of three IBM Firestone compute nodes based on the IBM POWER8 processors associated with NVIDIA K80 GPUs.

The second phase of the prototype, installed in December 2016, is composed of 12 IBM Minsky compute nodes with each containing:

  • 2 POWER8+ sockets, 10 cores, 8 threads per core (or 160 threads par node)
  • 128 GB of DDR4 memory (bandwidth > 9 GB/s per core)
  • 4 Nvidia new generation Pascal P100 GPUs, 16 GB of HBM2 memory
  • 4 NVLink interconnects (40GB/s of bi-directional bandwidth per interconnect); each GPU card is connected to a CPU with 2 NVLink interconnects and to another GPU with the remaining 2 interconnects (see figure below)
  • A Mellanox EDR IB CAPI interconnect (1 interconnect per node)

The tests on this prototype aim to quantify the work necessary for porting the applications on this architecture as well as the performance obtained. The preferred programming models are based on OpenMP 4.n or OpenACC directives; their maturity and performance will be evaluated.

At first, this platform was open to the porters of twenty selected codes, representative of different scientific domains.

Since 1 Avril 2017, Ouessant has been open to any community wishing to make tests on this platform. Requests for hours allocations must be made on the Edari site by presenting a project for preparatory access (“accés préparatoire”) on the Ouessant machine.

The IBM, Nvidia and Mellanox teams provide user support and participate in the workshops organised by this technology watch group.