From CPU to GPU in 80 days: Work Package 1

Palabos General
#1

Dear all

This post is to kick off Work Package 1 of the Palabos project From CPU to GPU in 80 days, which will keep us busy for the next two weeks. We suggest that any questions or remarks regarding the project are posted as replies to this thread.

The code developed in this work package will be pushed to the public Palabos fork of the project.

We will usually suggest community tasks at the beginning of each work package, but the project needs to get off the ground first. In the mean time, you may want to watch the recording of the project’s kick-off event or familiarize yourself with C++ Parallel Algorithms for LBM implementations on GPU.

The goal of WP1 is:
We implement five test cases in the original Palabos, including a built-in performance measurement framework. By the end of the project, some or all of the cases should run on GPU:

  1. Taylor-Green vortex [Uniform collision model, no boundary condition].
  2. Resolved flow in a porous media [Mesh-aligned inflow and outflow, bounce-back nodes].
  3. Multi-component flow segregation with pseudo-potential approach [Multi-phase coupling, no boundary condition].
  4. Flow around a sphere (no mesh refinement) [Off-lattice boundary condition around the obstacle, subgrid-scale model].
  5. Flow inside a tube (channel with circular cross-section) [Off-lattice boundary condition around the obstacle, subgrid-scale model].

Kind regards
Jonas

1 Like
#2

Hi Dr.Jonas
I have problem with how I can assign the inlet and outlet of my geometry in PALABOS.
I would appreciate your help

#3

At the end of WP 1, we implemented three of the five benchmark cases:

All of them have a built-in regression test and can run in benchmark mode or in production mode. Please go ahead and test them, don’t hesitate to post or email the results, which we will add to the spreadsheet.

We are still working on the cases with curved boundary conditions, which require some attention, as we are trying to simplify them in the hope to get them to run on GPU in a reasonable time frame.