BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20141118T223000Z
DTEND:20141118T230000Z
LOCATION:292
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Designing HPC systems for optimal application performance requires a tight integration of the compute elements, the interconnect and the storage system while minimizing cost and energy consumption. This presentation describes Bull’s approach to HPC systems architecture and illustrate the concepts on some recent installations.=0AThe compute nodes are based on a combination of CPUs or HPC accelerators depending on the application profile. The interconnect based on InfiniBand is configured with a compact fat-tree topology.  The storage nodes which aggregate a huge capacity (PBs) are carefully placed into the network fabric in order to provide fast data access with full bandwidth (100s GB/s) and high availability. For lower cost, the platform integrates compute nodes and parts of the interconnect. The direct liquid cooling technology minimizes the energy consumption required for cooling and thus reduces operating costs. =0AFinally, the key ideas guiding the current developments for Bull next generation system are explained.
SUMMARY:Practical considerations for HPC systems architecture
PRIORITY:3
END:VEVENT
END:VCALENDAR