|
||||||||||||||
Plenary
Lecture
Abstract: Due to processing speeds and memory
limitations of existing supercomputers, many current
simulations cannot faithfully simulate important
realistic phenomena. Thus these simulations are not
accurate enough to allow design and optimization of
important devices. In order to simulate realistic
situations very fine grids (e.g. on the order of tens of
billions of points) are sometimes needed, requiring
petascale computing systems. However, running existing
codes on bigger computers is not the answer. Fresh
designs are needed as well as implementation strategies
that take advantage of the main characteristics of
petascale architectures. For example, algorithms that
take advantage of multi-level parallelism and, within a
node of such an architecture, address the “memory wall”
aspect of multicore architectures where the cost of
arithmetic operations is much smaller than memory
references. One example of a problem that can benefit
petaflop computing is jet noise simulation. Jet noise is
an important issue due to increased commercial
air-traffic, penalty fees for noisier aircraft, and
future stringent noise regulations as well as military
operational requirements. Simulations of realistic
conditions requires tens of billions of grid points.
Examples of large-scale simulations for this problem
will be given and scalability studies will be shown for
up to 91,125 cores.
|
|
|||||||||||||
![]() |
![]() |
![]() |
||||||||||||
copyright - designed by WSEAS |