CUDA test: N-body simulation

Gravitational N-Body simulation of 32768 particles using CUDA in a (fully dedicated) nVIDIA GTS250. The simulation is in 3D but the video shows a top/down projection.

The program runs at ~240 Gflops, using a (brute force) N^2 (particle-particle) algorithm, computing 10^10 pair forces every second.

The particle representation was made using allegro libraries, and the video was created linking 1400 images (generated in 15 minutes) using avidemux.

Category:

Science & Technology

Tags:

• nbody
• N-body
• simulation
• particles
• spiral
• spiral waves

[TheodenN]

Is the gravitiy center a stationary dot like a real black hole or does it have a rotating effect added manualy like a vortex?

Altough black holes pull everything in every direction they cause an vortex effect. Is that the case in this video? Or is it added manually.

[jahkr]

@TheodenN

in this case the bulge, within which is the central black hole, is represented by a fixed mass distribution. I did not add any additional effect, all you see in the center is due to differential rotation. This video contains a very simple model of the galactic disk performance without short-range interactions to search for patterns of spiral density and many details of the actual behavior of a galaxy have been overlooked.

[rotgertesla]

Hi, interesting simulation, i like how the galaxy was stable for a long time, did you use any dark matter?

[jahkr]

Yes, the fixed halo and bulge can be considered, for the most part, like dark matter. Thanks for your comment

[TheReasonWhyGuy]

looks remarkably like a galaxy towards the end :)

[LSDDao]

=> this creations are very interesting, and even more if you could see them from different views/angles/etc. Moreover, I want to build some configuration like this one (galaxy-like). It's good to read here initial conditions, it's really helpful, and maby soon i'll get somethig like this. Actually my plans are vast and includes writting a raytracer, smoke-render and so on ;) But i think that i will have to buy new card that supports OpenCL, and of course that is more efficient :)

[LSDDao]

I like this one! Now i'm working in the same field. I have only my ati 3870, so i work with brook plus (it's doesn't support OpenCL). My algorithm is pretty ordinary (all-pairs). I've get about the same perfomance on my card (you said that in this video was 1400 images, generated in 15 minutes, thus it's 1.5 fps realtime) on 32786 particles. Now i want to use OpenGL to render particles (or maby do some calculations for "camera" rotation/moving in the rendered world), because exploring this =>

[jahkr]

Correction: For the mass, I used 70% for the halo, 15% for the bulge and 15% for the disk. sorry

[jahkr]

Making litle changes in the parameters is possible to get differents spiral patterns and get a central bar. By the other hand the number of particles is critical in the stability of the spiral structure, I have never obtained a stable spiral using less than 100K particles. You must understand that is only a model and it has many limitations (for example, the gas component is not included in the simulation)