Hello, i'm writing an application which requires heavy branching and the code runs faster on my cpu the on the 5870. I'm in the beginning of the development and I expect the performance will increase with smarter coding but I wanted to know: Since there is a new architecture on the southern island cards can I aspect a significant performance gain over my 5870 for algorithms depending on heavy branching?
Solved! Go to Solution.
Of course it depends on a lot of details but I believe it's fair to say that, yes, in general you can expect significant performance improvements. For one thing the architecture of the southern island gpus doesn't use the 5VLIW or 4VLIW architecture. Thus you can get fairly good instruction packing even if there is some level of branch divergence.
Are you sure you can't rework your algorithm to eliminate some of the branching?
I can and I will. But this will work only to a certain degree.
Of course it depends on a lot of details but I believe it's fair to say that, yes, in general you can expect significant performance improvements. For one thing the architecture of the southern island gpus doesn't use the 5VLIW or 4VLIW architecture. Thus you can get fairly good instruction packing even if there is some level of branch divergence.