Steady state solution of a diffusive process for given physical time steps using the pseudo-transient acceleration (using the so-called "dual-time" method).
💡 Use ParallelStencil.jl and ImplicitGlobalGrid.jl for the (multi-)XPU implementation. You are free to use either @parallel or @parallel_indices type of kernel definition.
What's all about. Brief overview about:
- the process
- the equations
- the aims
- ...
The methods to be used:
- spatial and temporal discretisation
- solution approach
- hardware
- ...
Results section
Report an animation of the 3D solution here and provide and concise description of the results. Unleash your creativity to enhance the visual output.
Briefly elaborate on performance measurement and assess whether you are compute or memory bound for the given physics on the targeted hardware.
Strong-scaling on CPU and GPU -> optimal "local" problem sizes.
Multi-GPU weak scaling
Provide a figure depicting convergence upon grid refinement; report the evolution of a value from the quantity you are diffusing for a specific location in the domain as function of numerical grid resolution. Potentially compare against analytical solution.
Provide a figure reporting on the solution behaviour as function of the solver's tolerance. Report the relative error versus a well-converged problem for various tolerance-levels.
Discuss and conclude on your results
Provide here refs if needed.