who schedules what?

Hey guys…

i have a small question . i read somewhere that whenever we launch a kernel , the SMA load the grid , then each SM gets a block.

i also read that SM schedules the warps within , my question who schedules the SM ? the SMA?


What does “SMA” stand for?

Blocks are distributed to the streaming multiprocessors (SMs) by something at the chip level (didn’t know it had a name). Each SM has an internal scheduler that decides which warp instructions from the block to run next, since that depends on which warps are waiting for memory reads to finish or on the results of previous instructions.

Thanks for the reply!

SMa stands for Streaming MultiProccesing Array , before blocks are distributed to the Sms… i saw somewhere written that the grids is distirbuted to the SMa