jetson_benchmarks icon indicating copy to clipboard operation
jetson_benchmarks copied to clipboard

Published 20 hours ago • verified publisher icon NVIDIA-AI-IOT

some confusion about the nx_benchmarks.csv file.

Open WangFengtu1996 opened this issue 4 years ago • 2 comments

HI, all I want to know the means about some auguments (like Devices BatchSizeGPU BatchSizeDLC WS_GPU WS_DLA input and output)

  • The BatchSizeGPU is the number of picture inferencing once?

Thanks for your time.

WangFengtu1996 avatar Jul 23 '20 11:07 WangFengtu1996

As the name suggest, batchsizeGPU is the default batch size for GPU wheareas batchsizeDLA is for DLA. WS is for WorkSpace this is TensorRT parameter which "controls the maximum amount of workspace that may be allocated" For more information

Please let me know if you have any question.

ak-nv avatar Jul 28 '20 18:07 ak-nv

Hi, I have some new questions.

  • How to be sure the proper value of batchsizeGPU, batchsizeDLA WS for different models?
  • How to understand these parameter to have an impact on the performance for NX?
  • what role s do DLA and GPU play in the benchmarking?
  • Is NX's performance is sum of DLC and GPU?

thanks for your time.

WangFengtu1996 avatar Aug 04 '20 12:08 WangFengtu1996