     |
 |
 |
 | | | | |
 |
|||
 |
||||||
          |
 |
 |
||||
|
Building a Million-Particle System Conclusion Of course processors are never fast enough, so up to now the implementation on the first generation of floating point GPUs simulates and renders a 1024×1024 texture of particles in real-time only with few effects and without sorting. Sharing the GPU with other techniques and using the full feature set currently allows up to 512×512 particles. The performance is expected to improve significantly with the upcoming generation of PC graphics hardware. This paper has shown how to design and implement a state-preserving physical particle simulation on current programmable graphics hardware. The simulation can use either Euler or Verlet integration to update the particle positions. Other particle attributes are simulated with less complex algorithms. Without permanent storage they are always evaluated on demand. Additionally, an efficient parallel sorting algorithm for particles has been introduced. The main strength of GPU-based particle systems is the low cost of individual operations on the data set. Once a basic algorithm is implemented, endless ideas for manipulating velocity and position come up and are easily implemented in higher level shading languages. New, yet unimplemented ideas include e.g. collision with arbitrary geometry and local forces being attached to the particles themselves. These forces basically lead to a second order particle system (cf. [Ilmonen2003]). Other topics for further discussion are the application of the algorithm to constraint PS (e.g. to simulate cloth or hair) and its modification for other rendering techniques instead of individual geometry. The only part of the particle simulation remaining on the CPU is the allocation and deallocation of particles as these do not map well to current parallel hardware. Future graphics hardware might support simple allocation algorithms with special-case serial registers that could be used as stack pointers. Further research should also be done on improving the exploitation of frame-to-frame coherence by the sorting algorithm. Currently, the full sorting sequence of the algorithm is divided evenly over several consecutive frames. Since the sorting result is not necessarily exact, it is possible that certain parts of the sorting sequence are visually more important than others and ought to be executed more often. How applicable to upcoming video game console hardware the introduced state-preserving particle simulation is, remains to be seen. But the trend towards increased multi-processing hardware is a good indication that the parallel computation of particle systems will grow in importance. Acknowledgments I would like to thank my colleagues at Massive Development, especially Ingo Frick, Dr. Christoph Luerig and Mark Novozhilov, and Prof. Andreas Kolb from the University of Siegen for the fruitful discussions and support in writing this paper. I also thank very much Sieggi Fleder for never giving up on my Germanic English. Furthermore, I am grateful to Matthias Wloka and his colleagues at NVIDIA, who helped the demo implementation to stay on the cutting edge of technology. References Batcher1968: Batcher, Kenneth E.; Sorting Networks and their Applications. In Spring Joint Computer Conference, AFIPS Proceedings 1968 Buck2003:
Buck, Ian; Data Parallel Computing on Graphics Hardware,
2003, Burg2000:
van der Burg, John; Building an Advanced Particle
System, Game Developer GPGPU2003:
Harris, Mark et al.; GPGPU Website, 2003-2004, http://www.gpgpu.org/ Harris2003:
Harris, Mark, Real-Time Cloud Simulation and Rendering,
Department of Ilmonen2003: Ilmonen, Tommi; Kontkanen, Janne; The Second Order Particle System. In WSCG Proceedings 2003 Jakobsen2001: Jakobsen, Thomas; Advanced Character Physics. In GDC Proceedings 2001 Lang2003: Lang, Hans W.; Odd-Even Merge Sort, 2003, http://www.iti.fhflensburg.de/lang/algorithmen/sortieren/oemen.htm Mark2003: Mark, William R.; Glanville, R. Steven; Akeley, Kurt; Kilgard, Mark J.; Cg: A System for Programming Graphics Hardware in a C-like Language. In SIGGRAPH Proceedings 2003 McAllister2000: McAllister, David K.; The Design of an API for Particle Systems, Technical Report, Department of Computer Science, University of North Carolina at Chapel Hill, 2000 Microsoft2002: Microsoft Corporation; DirectX9 SDK, 2002, http://msdn.microsoft.com/directx/ NVIDIA2001: NVIDIA Corporation; NVIDIA SDK, 2001-2003, http://developer.nvidia.com/ NVIDIA2002:
NVIDIA Corporation; OpenGL Extension NV_pixel_data_range,
2002, OpenGL2003:
OpenGL ARB; OpenGL Extension ARB_vertex_shader, 2003, Percy2003:
Percy, James; OpenGL Extensions, 2003, Purcell2003:
Purcell, Timothy J.; Donner, Craig; Cammarano, Mike;
Jensen, Henrik W.; Reeves1983: Reeves, William T.; Particle Systems – Technique for Modeling a Class of Fuzzy Objects. In SIGGRAPH Proceedings 1983 Sims1990: Sims, Karl; Particle Animation and Rendering Using Data Parallel Computation. In SIGGRAPH Proceedings 1990 Verlet1967: Verlet, Loup; Computer Experiments on Classical Fluids. I. Thermodynamical Properties of Lennard-Jones Molecules, Physical Review, 159/1967
|
||||||||||||||||||||||||||||||||||||||||||||||||
Features
|
|
|