Measuring different DRAM bandwidths for different applications, and both are below peak

[Anoop11]

Hi,

I'm comparing two programs, call them program A and program B. In program A, Intel Advisor shows a DRAM roof with bandwidth 39.98 GB/s. In program B, I see a DRAM roof with bandwidth 21.12 GB/s. My laptop has an Intel i7 11800H, which has a peak bandwidth of 51.2 GB/s. My first question is, why is Advisor measuring different bandwidths for different programs? Shouldn't this just be a characterization of my laptop, and not the program? Second, should these measured quantities be closer to the rated peak of 51.2 GB/s?

I attached screenshots of program A and program B below, titled program_A.png and program_B.png, respectively.

Program AProgram B

[Jennifer_D_Intel]

The difference has to do with the number of cores selected in the top left of the graph. The first screenshot with 39.98 has 8 cores selected, and the second has 1. The reason for these values is that the first application has 8 threads, and the second has 1 thread. A single CPU is going to have less available bandwidth than all of them.

Advisor calculates bandwidth roofs by running benchmarks prior to analysis. These may be lower than the reported maximums for the hardware.