CPU time and scalability

I was wondering about the typical CPU time associated with OpenLB. I will have to run a system 240x240x50 nodes, and am in the process of applying for cluster time, so I would appreciate it if you knew of any CPU benchmarks for 3D code for OpenLB.


From running OpenLB simulations with D3Q19 lattice on various modern CPU’s, I have observed that the performance ranges somewhere between 0.8 and 2 million site updates per second and per core, depending on the hardware and many other parameters. That’s of course only valid if you do simple BGK or MRT or similar on a regular domain. Under the same assumptions, I would expect the performance to scale well up to at least 100 cores in your example, after which scalability really depends on the type of machine you are using.

If you happen to measure the performance of your application at some point, don’t hesitate to post your findings here. It’s always interesting to have values for comparison.