FleX is a unified particle physics framework that can be used for all kinds of real-time simulations for all object types: rigid- and soft-bodies as well as fluids and gases. It utilizes the power of CUDA and NVIDIA GPUs and can do simulations with tens of thousands of particles in real-time. It's only natural to wonder whether FleX can produce results applicable not only for real-time, but also for high-precision simulation for animations and feature films.

Recently CGIcoffee.com blog author Armen Barsegyan conducted two studies involving NVIDIA FleX with two main goals: to see how FleX could be integrated into a Digital Content Creation package (in this case - Autodesk Softimage) and whether it could speed up such routine tasks as cloth and fluid simulation for production work.

In the first example FleX was used to simulate two highly dense cloth meshes in two scenarios: a flag waving in the wind and a rapidly twisting cloth banner. Both meshes of over a hundred thousand and 25K vertices respectively, converted to a corresponding number of interconnected and colliding particles, were successfully simulated with FleX directly from Softimage using API integration via a custom ICE-node. At 10 iterations and 20 sub-steps FleX was able to achieve 1.5-3 fps sim speed on a consumer-grade NVIDIA GPU with perfectly stable results good enough for production.

In his second study author gives a brief overview of various fluid simulation methods and tools and successfully challenges FleX library with a task of simulating fluid consisting of 500K to 1 million particles. As expected, FleX, as a framework based on a robust PBF method, is perfectly suitable for small to medium scale high-quality fluid simulations with an added bonus of supporting per-particle phase changes and a fast iteration speed thanks to the power of the NVIDIA GPU and CUDA.

Both studies are available at CGIcoffee.com:

At NVIDIA we are always excited to see the things that developers come up with using our APIs. We salute Armen's innovative use of FleX and the quality of his results.