Hello,
This article:
https://devblogs.nvidia.com/parallelforall/how-access-global-memory-efficiently-cuda-c-kernels/
Talks about coalesced memory access when reading from the global memory:
“The device can access global memory via 32-, 64-, or 128-byte transactions that are aligned to their size.”
Is there a CUDA API function that can return the byte size of the “transaction”?
If that number depends on the GPU version is there a table showing those values?
Thanks a lot,
M