2

I am just trying to fetch the TPU utilization details in coral dev board. But can't find any relevant for the same..!

Please share your findings / suggestions..!

Bhanuchander Udhayakumar
  • 1,581
  • 1
  • 12
  • 30
  • 1
    There are no performance registers to read, so taking benchmarks and timing measurements are your best way to estimate utilization. Look at the inference times after Figure 6, here: https://coral.ai/docs/dev-board/get-started/#6-run-a-model-using-the-pycoral-api This shows the difference between the PCIe bus transfer with inference, and just one inference time. Each model will have a different utilization depending on the number of layers and parameters. There are public benchmarks you can review: https://coral.ai/docs/edgetpu/benchmarks/ The models are available to run your own tests. – usbguru Dec 06 '20 at 00:11
  • @usbguru Thanks for your response.. Though i ve already came across those points.. opened this question to find out still any thing there.. i ve missed.. – Bhanuchander Udhayakumar Dec 07 '20 at 10:37
  • This is a good question. I've been looking through the `/sys/class/apex/apex_0`, and there is some stuff about temperatures there. Im yet to find TPU utilization though. – Isaac Padberg Mar 01 '22 at 19:36

0 Answers0