use nvprof --metrics to collect sm information following timeline order

Hi all,

I want to use nvprof --metrics to collect sm information like sm_efficiency by following the timeline order, but i don’t know
which command should i add here.


By “following the timeline order” - do you mean that kernels are profiled in the same order as they appear on the timeline?
For kernels within the same CUDA stream it is guaranteed that nvprof profiles the kernels in the same order. But across streams, order can change during the profiling run.
Not the kernel execution is serialized when metrics are collected.
Your requirement is not clear. Can you please provide more details?