Skip to content

TensorRT benchmarking parameters on T4 GPU #349

Open

Description

@altair199797

Dear Authors,

firstly, thank you for your great work!

I was wondering what were your benchmark parameters? I would guess you used trtexec for benchmarking?

Unfortunately, I can't reach your benchmarking numbers with this command on colab:
!trtexec --onnx=original.onnx --saveEngine=original.engine --shapes=images:1x3x640x640 --fp16 --noDataTransfers --useCudaGraph --builderOptimizationLevel=5 --maxAuxStreams=10 --useSpinWait --warmUp=20 --iterations=2000 --memPoolSize=workspace:7g

The lowest number I ever got, was 9.2ms (vs. 8.07 ms reported in the paper). I tried with and without maxAuxStreams and a lower --builderOptimizationLevel=5.

Could you give me some insight into this?

Kind Regards,
Moritz Nottebaum

Activity

  1. SebastianJanampa commented on Mar 28, 2026

    @SebastianJanampa

    Hi,
    You could try using the single-benchmark-artifact repo developed by Roboflow.

  2. altair199797 commented on Mar 28, 2026

    @altair199797
    Author

    Thanks for pointing out!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions