Shared memory can be accessed really fast, but it's only available to threads within the same thread block. Each lane in shared memory can give back a warp per block, so it's like you have 32 different memory banks that can all be accessed at the same time.
Shared memory can be accessed really fast, but it's only available to threads within the same thread block. Each lane in shared memory can give back a warp per block, so it's like you have 32 different memory banks that can all be accessed at the same time.