False Sharing in GPUs

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all:

Does false sharing problem exist in the GPU computing?

Thanks
malik

2

False sharing is not a CPU or GPU problem, it’s just a cache-related problem, so to be clear: YES, you could have false sharing on GPU.

Notice that many GPU use non-coherent L1 cache, and thus are not exposed to this problem, but instead to the problem of coherency when sharing data between SM or SMX or any logical sub-unit.