Hey, I’ve recently started developing program under OpenCL and created Kernel and it’s executed corretly, but somehow calculations made by my Intel CPU are faster then same one made by my grafic card NVIDIA GeForce 740. How is it posssible??? What could I possibly do wrong? Is there something specific about code optimization?
Here below is code of my Kernel, if accessing it is also needed I’ll attach it later.
[ATTACH=CONFIG]72[/ATTACH]
That kernel looks like it was code-generated, not hand coded. In any case, one source of slowdown is that each work item reads 16 doubles from global memory. While they can be broadcast within each workgroup, lots of workgroups will be duplicating the reads over and over. You’d be much better off passing these as a double16 parameter, which uses constant memory and will be faster. Secondly, you’re using doubles, which perhaps aren’t fast on your GPU. If you can use float instead, they will be faster.