@erik-nl: you are right. The GPU version don't work correctly with low-res mesh. It is for this reason I kept the CPU simulation. It is a problem which exists and is very difficult to solve when a function updates competitive elements in a system highly parallel. This problem comes with the springs computation where the position of two nodes is updated in the same function. I want to improve this part of the GPU simulation but in a similar case, the result will always be worse than the CPU simulation. Please, make a try with a character and a cloth or hair, You will see the result is very good and very fast.

____________________________________________

Follow me on Twitter : @VWDynamics

Watch demo videos on Youtube