No performance gain after 0 bank conflicts

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi All,

I removed bank conflicts in my program and I’m checking that through CUT_BANK_CHECKER which reports 0 bank conflicts in my kernel code. But I got no performance gain after that !!

Has any anyone encountered this before or have any clue what’s going on?
I would appreciate any help, I’m kind of stuck with this issue …

Thanks Much.

2

You’re probably just so constrained by some other aspect of your code (I’m betting global memory bandwidth) that you get no gain by removing bank conflicts. Are your memory accesses coalesced?