Hello
I want to know valuse about DRAM such as
dram__bytes_read.sum.per_second
but I can’t find the value on the result of profiling (Figure below).
How can I get the values(dram__bytes_read.sum.per_second) for the DRAM?
Also, I want to know number of GPU clock cycles required by the execution of one CUDA kernel.
Can I get the value( number of GPU clock cycles required by the execution of one CUDA kernel) through profiling GPU?
If so, what value do I have to check?
When specifying individual metrics to ncu with the --metrics switch, they will show in the UI Details page as a new, separate cmdline section. You can also view all metrics collected in the current profile result on the Raw page where you can filter for individual names, too.
Also, I want to know number of GPU clock cycles required by the execution of one CUDA kernel.
You can refer to the Elapsed Cycles entry on the GPU Speed Of Light section. The respective metric is called gpc__cycles_elapsed.max
Which exact version of Nsight Compute are you using to collect the data, and which one to view it in the UI? From your screenshot, I think you might be using a rather old UI version with a newer report, due to which some of the data is not shown properly. I recommend to the use the same version for both collecting and viewing the report.
