I am working with particles and have achieved 100x speed up when I use coalescing/shared memory, but would like to compare this performance with a different code using textures, and then develop a hybrid if there is potential benefit. I am looking to use two arrays to store particle numbers of interacting particles. These, as well as the location and velocity of each particle, will probably change at each timestep. Do I need to bind/unbind at each timestep, or bind once at the beginning of the code and unbind once at the end? If binding/unbinding at each timestep when is the optimal time to do this?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| cudaBindTexture | 4 | 2448 | August 10, 2010 | |
| dynamic update of texture in a kernel is it worth it | 4 | 1629 | May 11, 2009 | |
| whats the execution time for cudaBindTexture() and cudaUnbindTexture()? | 0 | 3445 | August 3, 2009 | |
| texture memory binding performance | 5 | 2693 | June 13, 2012 | |
| Linear Memory Vs CUDA Array for texture binding | 2 | 17736 | January 21, 2010 | |
| What happens when you bind a texture? Under the hood I mean. | 6 | 2279 | May 5, 2009 | |
| Texture bind reuse or rebind | 4 | 623 | June 22, 2022 | |
| Sending the textures once to the graphics card. CUDA should be able to access it. | 1 | 3334 | April 10, 2008 | |
| How to do this (and be fast) | 7 | 7971 | February 9, 2008 | |
| texture binding and stream | 3 | 2064 | January 11, 2012 |