tensorflow-on-raspberry-pi icon indicating copy to clipboard operation
tensorflow-on-raspberry-pi copied to clipboard

Published 20 hours ago • verified publisher icon samjabrahams

tensorflow only occupies one CPU core out of 4 cores on Pi 3

Open kranthisai opened this issue 8 years ago • 4 comments

Hi,

I tried object detection model in Python. It took around 6 seconds for each image to process. I see tensorflow is using only one core once session starts, however pillow library for visualization uses all the cores for drawing bounding boxes on images.. I tried setting the intra_op_parallelism_threads, but no change..

I tried running same on Intel NUC, I can clearly see all cpus are being used.

kranthisai avatar Aug 04 '17 00:08 kranthisai

Does/will TensorFlow use multiple cores on ARM SBCs, e.g. nanopi m3 as well as Raspberry Pi?

hhemken avatar Oct 24 '17 22:10 hhemken

I have same problem. Only 1 core is being used... Can this be solved?

nan4k7 avatar Apr 06 '18 17:04 nan4k7

Did any of you ever get any further info about this? I also noticed tflite seems to be running on a single core..

GPhilo avatar Apr 26 '18 12:04 GPhilo

I have also experienced the same problem on an octa-core ARM CPU. Only one core is utilized by the tflite model!

zanazakaryaie avatar Mar 09 '20 07:03 zanazakaryaie