This week I spent more time trying to figure out some reference videos that I can use as comparison criteria, and more/so solidifying what I want to simulate. Although my plans for the project include multi-phase flow of other materials, right now just being able to melt ice would be a primary objective, so these are some videos of what I want to be able to simulate.
Here are some ref: videos for other types of highly viscous flow if molten lava.
Also, after looking at a few SPH base simulators, I decided to start out with the CUDA SPH implementation which is found in the CUDA Particles samples in the CUDA SDK. It's very well implemented, and seems the most well documented in comparison to some of the other open source simulators I've looked at. I spent a little bit of time playing around with it, and it seems incredibly fast and efficient at running the simulation. Surface extraction and rendering is a different issue and will definitely slow down the process to some degree, but that part will come much later.
Terry this was a great post, I think the biggest thing is to think back to your constraint dynamics. Start to list all the properties you want your particles to have, the forces, and the constraints you want to enforce.
ReplyDeleteThen look at the reference videos and list these aspects. So what is acting on ice: Gravity, the center of mass is changing, ice expands when it freezes so its losing volume ... etc... for the videos
Terry, glad to hear you are using CUDA for your senior design project. Although I haven't looked at the SDK example myself, I'm sure it is "incredibly fast and efficient at running the simulation" as you observed. When you have a data-parallel algorithm like this, a naive GPU implementation can be 10x faster than a CPU implementation, and a tuned GPU implementation can be 100x.
ReplyDeleteInteresting point pat, one option for the 'tuned' code would be to start from this base http://graphics.ewha.ac.kr/gSaP/
ReplyDeleteinstead of the Green SDK example.