2D CFD explosion simulation NVIDIA CUDA

~130 000 particles gas explosion simulation. Coloring based on velocity.

GPU used: GF 8400M GS, 16 cores @ 400 mhz with 128 MB dedicated memory @ 400 mhz.

This is slightly modified sample from CUDA SDK. I have changed all algorithms from 3d to 2d, thanks to that, now program use 50% less memory and is 30% faster (maximum ~800 000 - 900 000 particles with 0.5 - 1 fps, and 32768 particles @ 35-40 fps). Why 2D? Because 3D with 350 000 partciles were giving too low resolution and accuracy. Even 130 000 particles in 2D gives much more realistic effects.

Second change was dynamic coloring based on particle's velocity (blue high speed, white low speed). For now its linear scale, based on average velocity at specified frame. In future I plan to change it to constant logarithmic scale. Unfortunatelly at the moment coloring slows down rendering by about 10-20%

Last change was adding diffusion effect (before calculating every frame adding smal random velocity vector).

I think effect is prety good :) More changes are on the way :)

NVIDIA CUDA site
http://www.nvidia.com/object/cuda_home.html

Category:

Science & Technology

Tags:

• cfd
• explosion
• nvidia
• cuda
• fluid
• gas
• simulation
• mechanika
• płynów

[dalcimar]

In documentation says that CUDA programs require at least 256 MB of memory. You use only 128 MB correct?

You had any problem due the low memory? Any problem in implementation or instalation of driver?

[B4ndItOo]

yes, i have only 128mb. Generally speaking, there were no major problems.

[seleneium]

Hmm... out of curiousity what program did you use to utilise Cuda, I know matlab is one..

I guess the question is how were you able to set up the simulation (I have a 9600m gt so I am interested in conducting these and similar simulations)

[B4ndItOo]

i was using visual studio 2005 c/c++. Generally i was playing with sample projects that come with cuda sdk

[seleneium]

interesting I did not know that the 8400m gs had Cuda

[B4ndItOo]

every gf8xxx and newer support both Cuda and physx. Some time ago there was problem with drivers for mobile gpu's, but recently nvidia released official drivers that cuda.

[cskjain]

Any suggestions for learning CUDA?? (for beginners plz)

[nvidiacuda]

Are you releasing the code? You should also consider posting your app to CUDA Zone. Applications with video get a CUDA Zone T-Shirt and entered into a drawing for a GTX260