What does the metric "pcie throughput" mean specifically?

I would like to know if my kernel is high throughput completely.
Can you explain the meaning of “This value includes protocol overhead” in the context of pcie__read_bytes and pcie__write_bytes?
what does the term “protocol” refer to here?
Is the meaning represented by pcie__read_bytes and pcie__write_bytes the same in Nsight Systems, considering that both metrics exist within it?
Thank you.

Thanks for posting your question. I’m doing some internal investigation and will let you know what I find out soon.

There is some useful background on the PCIe protocol here.

Protocol overhead is data required for the transfer that isn’t the explicit payload. For example, some communication bits, etc… The metrics measure the same thing in Nsight Compute and Nsight Systems.

Thank you for your reply. I would like to ask further, is protocol overhead referring to the header of TLP, including or excluding the header of the data link layer?

It should include any data that goes across the PCIe bus, which would include any headers sent across the bus.

Thank you.
I have no further questions.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.