Hi, we are planning to build a CUDA simulation for interactive real-time analysis of many-particle-advections in unsteady wind fields. Likely bottleneck will be moving a lot of data (3D wind filds over many time-steps) as quickly as possible from disk to GPU memory. Possibly also the interpolation of winds to particle positions, some turbulence parameterization and high numbers of particles. I am wondering now which hardware setup would be good to have - in particular is there a way to transfer data from disc to GPU without having to go via the CPU?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Is CUDA right for me? (FDTD) FDTD user needs fast computations while handling massive 3-D arrays | 17 | 22538 | December 29, 2008 | |
| Can particle effects be created with CUDA? | 11 | 11472 | October 8, 2007 | |
| Stream Data | 1 | 1610 | June 9, 2011 | |
| Rendering data | 2 | 1485 | June 29, 2009 | |
| faster copying to gpu | 1 | 2434 | January 31, 2008 | |
| Two Questions! | 3 | 5295 | December 3, 2007 | |
| Help on fixing some poor performances (rookie) | 10 | 7231 | November 28, 2007 | |
| Particle Engine with CUDA Is there a way to avoid transfering the particles back to CPU Memory? | 4 | 8571 | March 23, 2010 | |
| Fluid Dynamics on GPU | 2 | 3735 | November 3, 2009 | |
| Need an advice about using CUDA for a certain task | 12 | 2008 | July 7, 2010 |