@lritter a few mins. obvs if you crank it you can quite easily get to a 100Gb cache overflow. Apparently if you double res, you get the 8X for volume, and also 2x for time. Taichi uses clever quantization/ sparsity to save speed/memory, but not sure what this particular implementation uses. but yeah, no longer than normal fluid sims in Blender really, res for res.