GPU Lattice Boltzmann fluid simulation with Sailfish

[incasmac]

Very interesting movie!!! If I ask a question of novice ^^!!! I work on a LBM solver, 3D with transport of a scalar (I solved the energy equation for the story ^ ^). I run the solver in parallel with MPI libraries. I wanted to know if it was feasible to transform the code for creates GPU solver? (I use FORTRAN90) thx

[elenglad]

It would be nice if someone left any information about the music used in this video.

[boogie0815]

What exactly is 1 LUP - is it calculating all of the in/outgoing of one grid cell or is it updating only one direction (=9 LUPs is one grid cell at D2Q9)?

[ytmjanusz]

@boogie0815 LUPS stands for Lattice-site Update Per Second, and it does not distinguish directions in any way. For instance, if the simulation domain is a 100 x 100 grid, and 10 time steps can be calculated per second, the performance estimate is 100*100*10 = 100,000 LUPS = 0.1 MLUPS.

[boogie0815]

@ytmjanusz Alright, got it. Thx for quick reply. I am working on a GLSL version of the LBM algorithm. Despite mine is _very_ simple, I have less than 100 MLUPS on a HD4770 :(

Are you still working on it? Project seems a little abandoned on the net - I would like to see some D3Q19 stuff ;)

[ytmjanusz]

@boogie0815 The project is very much alive and I'm still actively working on it. I have some patches queued up, they just haven't made it to the main repository yet. Also, D3Q19 has been supported for a long time now, there are just no YouTube videos for it ;)

[PeterMacGonagan]

Great! I would like to compute the lift on an airfoil. Do you think it's possible with your software?

[ytmjanusz]

The current version of Sailfish doesn't have any support for calculating drag/lift forces (that being said, it should be a relatively easy thing to add..). A more limiting factor might be the stability of the simulation in high-Re flows. What ranges of the Reynolds number are you interested in?