Data transfer between DLA and system memory

Hello there, is there a way to get the memory transfer metrics between DLA and the system memory?

Thanks in advance.

Hi,

Do you want to use DLA with TensorRT?
If yes, do you want a bandwidth test for GPU ↔ CPU memory?

/usr/local/cuda-10.2/samples/1_Utilities/bandwidthTest

Thanks.

Yes, I use DLA with TensorRT, so I want to capture the system memory read/write bytes caused by DLA.

I saw the bandwidthTest file that you provided, where device means GPU (not equal to DLA) and host means CPU, no clues on how to test for DLA <-> CPU.

Therefore, my only question is:

How to get memory read/write bytes for DLA <-> CPU?

Hi,

Since only TensorRT can deploy a task on DLA, we don’t have a sample that only measure the bandwidth.
Sorry for the inconvenience.

Thanks.

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