Global memory access for float and integer, the speed is the same, right ?

[Michael_H1]

Global memory access for float and integer, the speed is the same, right ?

Can the memory coalescing happen in the same manner for integer values ?
or float type is the choice for gpu ?

Thank you

[njuffa]

The hardware has support for 32-bit, 64-bit, and 128-bit aligned loads. In CUDA, the C/C++ data types ‘float’ and ‘int’ both comprise 32 bits, so if the access pattern is the same, the same amount of coalescing takes place, and the memory throughput is the same.

[Michael_H1]

The hardware has support for 32-bit, 64-bit, and 128-bit aligned loads. In CUDA, the C/C++ data types ‘float’ and ‘int’ both comprise 32 bits, so if the access pattern is the same, the same amount of coalescing takes place, and the memory throughput is the same.

Thanks !