Workshop on Modern Scientific Computing
We invite all those interested in modern scientific computing trends to a free 2-day hands-on workshop taking place March 2-3, 2016 from 9:00-17:00. Due to limited spaces sign up as soon as possible by sending an email to organizer Allan P. Engsig-Karup (See below).
How To Scale Scientific Applications from Laptops to Supercomputers with PETSc?
What is the common link between weather forecasts, shape optimization of airplanes, and faster and more power-efficient transistors?
All three can be modeled and thus better understood through partial differential equations (PDEs). Analytical solutions of PDEs are usually impossible to compute for practical applications, hence scientists and engineers rely on numerical solution methods. The PETSc library provides a rich toolbox for computing these numerical solutions on a variety of machines ranging from low-end laptops to the largest supercomputers.
Finding the best numerical solution method for a given application is challenging. Who wants to wait hours or days for each simulation run if the best solver can find a solution within minutes? Unfortunately, there is no silver bullet recipe for designing such an optimal solver. However, there are established best practices and experience on how to tackle certain classes of problems most efficiently. In this workshop you will learn about these best practices and gain hands-on experience on how PETSc can help you to find fast solvers.
During the workshop you will acquire and expand an inter-disciplinary skill set:
- you will get a better understanding of how linear and nonlinear solvers work and how they can be accelerated,
- you will see how physical properties such as diffusion can be leveraged for the design of faster solvers,
- you will run solvers on parallel machines and scale them from a single process to hundreds of processes,
- you will learn efficient workflows with debugging and profiling tools inside and outside of PETSc.
- Assoc. Prof. Allan P. Engsig-Karup, apek@dtu.dk
- Assoc. Prof. Niels Aage, naage@mek.dtu.dk
Responsible:
The seminar is organized by:
Material
