 |
||
|
||
Home > TERATEC FORUM > Workshop 6
Algorithms and parallel software
Chaired by Guillaume COLIN DE VERDIERE, CEA
The session on « Algorithms and parallel software » was the opportunity to present the current activities to use efficiently the supercomputers that are available to industrials, academics and researchers. A number of themes have been selected to illustrate the variety of the subjects, especially in the light of the exaflop race.
Modern supercomputers are more and more complex. This complexity must be understood and mastered in a constraint budget. Application performance modelling becomes a key step to understand what could be future computer architectures and choose between potential options. In order to mask part of the difficulties to programmers, runtimes are components that are taking more and more importance with each machine generations.
Machines are not the only one to face increased complexity. Real life codes are difficult to develop because they have to solve multiple physics problem simultaneously (tight coupling), introducing specific constraints and challenges to reach the maximal performance of a modern supercomputer. To help codes (developers), it is important to specifically develop libraries keeping performance in mind.
Taking a significant part of the complexity of current and future computers, the hardware in use plays a significant role in the service to the users. The TOP500 shows that we can’t neglect the impact of accelerators on scientific computations. Last, the budget pressure that we all face as depicted through the GREEN500, forces studying alternative solutions taking energy constraints into account.
The talks of this session was cast light on each of those issues.
|
|
|
|
|
|
Abstract & Biography 
 |
 |
 |
 |
 |
 |
 |
 |
