Hi,
I use ncu to profile my kernel and want to see the roofline. It comes out with a roofline chart but it didn’t show the achieved value for my kernel?
Is it because my FLOPS is too small?
Actually, my kernel is about graph algorithm’s acceleration. Most of the operations are integer operations instead of floating points.
Which metric should I use if I want to get the achieved value through ncu for my kernel?
Can you please provide the report or a minimal reproducible with NCU command line, NCU version, OS, NVIDIA driver and GPU.
Are you sure you have floating point operations in the kernel?
I do not have FP operations in kernel. Only integer operations like memory access operations. Can we get the achieved value in this case? I mean it’s ok if we do not have FLOPS. But should have similar metrics to the achieved performance?
Command line just use ncu --set full -f -o xxx.
NCU does not currently have a roofline for integer operations (e.g. IMAD, IMUL, IADD*, ISCADD, LEA, etc.)
