LFD-A-Light-and-Fast-Detector icon indicating copy to clipboard operation
LFD-A-Light-and-Fast-Detector copied to clipboard

Published 20 hours ago • verified publisher icon YonghaoHe

why trt mode consumed more GPU and MEM space？

Open chainfitness opened this issue 2 years ago • 2 comments

thx this awesome work! I just want to deploy this awesome work in practice, but i find that trt mode consume more GPU and MEM space.The GPU refered above is Tesla T4, i'd appreciate if the prompt repley.

chainfitness avatar Mar 02 '22 02:03 chainfitness

@chainfitness We do not encounter your case before. Would you provide more information?

YonghaoHe avatar Mar 03 '22 11:03 YonghaoHe

I'm obilged for quikly reply.I want to deploy face detection from this work to nvidia tx platform. But the official installing TensorRT version from TX is 8.0.1.6,which is too new to run predict_tensorrt.py. So,i move code to Tesla T4 that installed TensorRT 7.2.2.3, it works! But i find that trt mode consumed more GPU and MEM space, GPU and MEM consumption running predict.py was 977M , 2123M respectively,however, running predict_tensorrt.py was 1486M and 3714M. The pretrained weight was downloaded from image predict_tensorrt.py is only modified the path of pretrained weight.

chainfitness avatar Mar 04 '22 02:03 chainfitness