Block Size Mode
Hi all,
I've made a few (very simple) tests changing the Block Size Mode (-blocksize parameter) and I've found that indeed there is diference in the simulation running time but not in the way I expected. From worst to best I've got:
1) Occupancy calculator mode (-blocksize:0)
2) Fixed block mode (-blocksize:1)
3) Empirical mode (-blocksize:2)
The difference wasn't huge but it surprised me because I believed occupancy calculator would be the best.
Is there any specific case in which each one of the modes would be more advantageous than the other?
The case I've tested was the Periodicity one with a GPU with CUDA capability 3.5 (GTX770), with the default resolution and twice that (0.002 and 0.001).
I've made a few (very simple) tests changing the Block Size Mode (-blocksize parameter) and I've found that indeed there is diference in the simulation running time but not in the way I expected. From worst to best I've got:
1) Occupancy calculator mode (-blocksize:0)
2) Fixed block mode (-blocksize:1)
3) Empirical mode (-blocksize:2)
The difference wasn't huge but it surprised me because I believed occupancy calculator would be the best.
Is there any specific case in which each one of the modes would be more advantageous than the other?
The case I've tested was the Periodicity one with a GPU with CUDA capability 3.5 (GTX770), with the default resolution and twice that (0.002 and 0.001).