CUDA. Bank conflicts in shared memory (device parameters)

Question

To avoid bank conflicts in sheared memory I need to know the number of banks and the bank size. As far as I know, for devices with compute capability 1.1 - 1.3 it's 16 banks with 32-bit element size. I have Tesla C2075 (compute capability 2.0). Do I have the same values and where can I find this? (I haven't found this in "cudaDeviceProp" information). By the way, I would be thankful if somebody gives me a link to some doc with all technical information about my CUDA device. One more think, I need to work with doubles, so I really hope, the bank size is 64 bit. If it's not, what kind of bank conflicts should I expect?

brano · Accepted Answer · 2012-10-18 09:18:29Z

3

The information you are looking for is in the cuda programming guide.
Appendix F.4.3 for cc 2.0.

answered Oct 18, 2012 at 9:18

brano

2,89421 silver badges15 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

CUDA. Bank conflicts in shared memory (device parameters)

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related