MSc Researcher and VFX Artist Alexandre Sirois-Vigneux has recently shared a short demo showcasing an impressive custom-made GPU-accelerated MPM solver with an embedded neural network, written in NVIDIA's parallel computing platform and application programming interface CUDA.
According to the developer, the neural network allows to "activate/deactivate parts of the simulation on the fly", making it faster. Using the solver, the creator produced an amazing exploding watermelon simulation, which took an hour to render on a GTX 1080 Ti GPU, planning to share more results in the coming weeks.
