medium10 pts · first answer countsgpu-computingcudaperformance

Memory coalescing

In a CUDA kernel, why is it faster when the 32 threads of a warp read 32 consecutive floats than when each reads a value far apart?

A.Far-apart reads are blocked by the GPU driverB.Neighbouring reads merge into a few transactionsC.Consecutive reads skip the GPU memoryD.Far-apart reads need a CPU round trip

Discussion 0

Join free to post in the discussion.

Memory coalescing

Discussion 0

Related questions