Intel VTune LLC Miss Count 0 Problem

[SrcJimmy]

Hi!

I have been using VTune for a few weeks and encountered this problem. My application uses the DRAM because it reads lots of memory greater than the LLC size and the VTune says the same, the DRAM average bandwidth is several GB/s. However, when I check the LLC misses which are the DRAM access it says 0. It doesn't make sense to me because if the DRAM channel is being used must be because the LLC has missed.

I have attached one of the experiments.

I hope someone can help.

Sincerely,

Jimmy

[yuzhang3_intel]

Which VTune version did you use? And can you capture uarch-exploration data as well?

[SrcJimmy]

Thanks for your response.

I am using the latest version (2025.0.0, build 629072)

Here is one execution with the uarch collection.

[yuzhang3_intel]

It looks like the result is aligned b/w uarch and macc, the bottleneck is the store bound.

uarch:

macc: