Future of GPU support in Espresso

[Rudolf Weeber]

Dear Espresso users,

many thanks to the participants of the GPU survey.
Your answers gave us an idea, how the GPU features are used outside the group 
in Stuttgart.

According to the replies, the most important GPU feature is the 
lattice-Boltzmann implementation. Uses include quite dilute systems, where the 
LB performance dominates the total simulation time.

We have decided to proceed as follows:

* The next feature release (4.2) will be made, before transitioning to the 
Walberla LB implementation. It will have GPU support including the existing LB 
GPU implementation.

* This release will receive bug fixes until the feature release thereafter, 
Espresso 4.3 is made. In the past, feature releases have been made in intervals 
of 12-18 months, but this depends on what features are contributed over a given 
period.

* After 4.2 is released, we will transition to the Walberla-based 
lattice-Boltzmann method in the development branch. This will be the CPU 
version, initially. We will then publish performance comparisons against the 
GPU LB in Espresso 4.2.. Based on these, and your answers to the survey, we can 
then figure out whether putting in the work to integrate Walberla's GPU support 
in Espresso and continuing to maintain our GPU build and test infrastructure 
makes sense.

Regards, Rudolf

-- 
Dr. Rudolf Weeber
Institute for Computational Physics
Universität Stuttgart
Allmandring 3
70569 Stuttgart
Germany
Phone: +49(0)711/685-67717
Email: weeber@icp.uni-stuttgart.de
http://www.icp.uni-stuttgart.de/~icp/Rudolf_Weeber