medium10 pts · first answer countsgpu-computingcudaperformance

Low occupancy in a kernel

Nsight Compute reports low occupancy for your kernel because each thread uses many registers. Why can low occupancy hurt speed?

A.The GPU clock speed drops with occupancyB.Each thread gets less global memoryC.Kernels with low occupancy run on the CPUD.Fewer warps are ready to hide memory latency

Discussion 0

Join free to post in the discussion.

Low occupancy in a kernel

Discussion 0

Related questions