This is a simple demonstration of the parralel Barnes-Hut treecode algorithm on 32 GPUs.
Two galaxies are interacting with each o This is a simple demonstration of the parralel Barnes-Hut treecode algorithm on 32 GPUs.
Two galaxies are interacting with each o This is a simple demonstration of the parralel Barnes-Hut treecode algorithm on 32 GPUs.
Two galaxies are interacting with each other
and collide. Two galaxies are modeled by 33,554,430 particles(4,272,458 for disk, 27,093,640 for halo, 2,188,332 for bulge).
Only the disk and bulge were visualized and the halo was invisible in this simulation result of movie.
--- Demonstration Environment ---
GPU: Nvidia GeForce8800GTS(G92)
GPU kernel: cunbody-1.9 library (http://progrape.jp/cs/)
N-body algorithm: O(N logN) Barnes-Hut treecode algorithm modified for parallel GPUs. The code was based on pC++tree by Prof. J. Makino.
Initial Cond. of data: 33,554,430 particles generated from the GalactICS package
MPI library: mpich2 1.0.7rc1
C++ compiler: gcc 4.1.2
Design languages: C++ and CUDA
CUDA version: CUDA 1.1
OS: CentOS linux version 5.1 (x86_64)
Place: Nagasaki, Japan
LastUpdate: 13 May, 2008
Copyright(c) 2008 by T. Hamada & K. Nitadori
Channel: Science & Technology Uploaded: May 21, 2008 at 9:15 am Author:cunbody1
Website is down. By the way, was done in real time or not?
amenzix(Saturday 6th of December 2008 08:27:51 AM)
Probably not, even with the 32 GPU's, I don't think that 33 million particles can be done in real-time.
cunbody1(Thursday 18th of December 2008 09:15:53 AM)
1.2 seconds per a timestep with 33,554,430 particles, which includes all of operation time for MPI communication, space decomposition, tree-construction, tree-traversal, time-integration and force calculation etc.
Although the opening parameter was implemented as a changeable parameter, this demo used it as 0.75.
Did you understand ?
And sorry for my late response.
