PPKE ITK PhD and MPhil Thesis Classes

2.3 Computational Fluid Flow Simulation on Body Fitted Mesh Geometry
with IBM Cell Broadband Engine and FPGA Architecture 53
Table 2.3: Comparison of different CNN ocean model implementations: 2GHz
CORE 2 DUO processor, Emulated Digital CNN running on Cell processors
CNN Implementations
Parameters Core 2 Duo CELL (8 SPEs)
Iteration time ( ms) 8.2 1.11
Computation time of a
354.2 47.52
72 hour simulation (s)
Power (W) 65 85
Area (mm 2 ) 143 253
(CNN cell array size: 128×128, 1 iteration)
2.3 Computational Fluid Flow Simulation on Body
Fitted Mesh Geometry with IBM Cell Broadband
Engine and FPGA Architecture
2.3.1 Introduction
In the paper of Kocsárdi et. al the authors applied the finite volume Lax-Friedrich
[53, 54] scheme for solving 2D Euler equations over uniformly spaced rectangular
meshes [55]. However, most real life applications of CFD require handling more
complex geometries, bounded by curved surfaces. A popular and often an efficient
solution to this problem is to perform the computation over non-uniform, logically
structured grids. Technically, this idea can be exploited either by employing
body fitted grids or by performing the computation in a curvilinear coordinate
frame following the curvature of the boundaries. Although in the latter approach
the standard 2D scheme over Cartesian geometry can be put to work, it is computationally
much more demanding than the former one, due to the expensive
operations related to coordinate transformation.
There is a number of reasons why we have chosen the IBM Cell multiprocessor
system as the restricted (bounded) architecture, namely, its high computing
performance, the double-precision floating point number support, the support of
the standard multiprocessing libraries, such as OpenMP or MPI and the freely
available software development kit.