Confused about the Orin GPU info

image

Isn’t Orin’s bandwidth is 204GB/s? Why is Nsight Profile showing 77.49GiB/s here?

Hi,

Do you get this by profiling your application?
This might be profiled with a custom app rather than the spec value.

Thanks.

Hi,i get above info by following command:

nsys profile -s none  -t cuda,osrt,nvtx --cuda-memory-usage=true --gpu-metrics-device=all  -o gridsample -w true /usr/src/tensorrt/bin/trtexec --onnx=../xxx.onnx --fp16 --allowGPUFallback --iterations=1000 --useCudaGraph=enabled

And I use different onnx models, the “Memory Bandwidth” is always 77.49 GiB/s, which confuses me.

Hi,

The info comes from the below TARGET_INFO_GPU table
https://docs.nvidia.com/nsight-systems/UserGuide/index.html#available-export-formats

CREATE TABLE TARGET_INFO_GPU (
    ...
    totalMemory                 INTEGER,                               -- Total amount of memory on the device (B).
    memoryBandwidth             INTEGER,                               -- Amount of memory transferred (B).
    clockRate                   INTEGER,                               -- Clock frequency (Hz).
    ...
);

We are checking this with our internal team.
Will update more with you.

Thanks.