Oppgaven er ikke lenger tilgjengelig

Use of many-core architecture for heart simulations

Simulating the heart is a well-known computational challenge. In addition to advanced mathematical models and numerical methods, software implementations must be able to effectively use modern parallel hardware. The many-core architure, represented by Intel's Xeon Phi coprocessor, is an important target platform.

The Xeon Phi coprocessor from Intel is the leading representative of the many-core architecture

The master-degree candidate is expected to carry out research on three particular subjects:

  • Manual and automated vectorization techniques that can unleash Xeon Phi's 512-bit SIMD capability per hardware thread
  • Effective design and traversal of data structures, in connection with using all of Xeon Phi's around 60 cores
  • The simultaneous use of both Xeon Phis and multi-core CPUs to enable heterogeneous computing

As an experimental testbed for the intended implementation innovations, a monodomain simulator of the heart will be ported to the Xeon Phi architecture. The underlying numerical method consists of cell-centered finite volume discretization on 3D unstructured tetrahedral meshes that represent the realistic geometries in a heart.

The learning outcomes include cutting-knowledge of scientific computing with a special focus on heart simulation, hands-on experience with programming of the modern Xeon Phi coprocessors, as well as expertise on heterogeneous supercomputing.

Emneord: heterogeneous computing, Xeon Phi coprocessor, SIMD
Publisert 25. aug. 2015 11:02 - Sist endret 29. aug. 2016 13:09

Omfang (studiepoeng)