Speaker
marcello pivanti
(INFN and University of Ferrara)
Description
This poster covers in details the architecture of the application-driven
AuroraScience supercomputer. This machine is a large 3D array of latest
generation Intel processors, with nearest-neigbor processors connected by
a network with toroidal topology. Each node has a peak (double precision)
performance of 100-150 Gflops. This machine has been engineered to be
reliably assembled in thousands of nodes, making it possible to deploy
systems with a peak performance of several hundred Tflops. This poster
complements a companion talk, presented by L. Scorzato, focusing on the
use of this system as an LQCD-optimized number cruncher. Here, we focus on
the structure of the machine and -- in greater details -- on the toroidal
interconnection system.
| Please, insert your presentation type (talk, poster) | poster |
|---|
Primary author
marcello pivanti
(INFN and University of Ferrara)
