Some questions on DeepStack GPU on Windows (Google Coral, settings, not using nVidia GPU...)


I installed DeepStack GPU version on Windows 10 with lastest nVidia drivers/cuda/cudnn.
I want to use it with BlueIris.
I have a Zotac ECM73070C with a i7-10700 and RTX3070.
This hardware is only for BlueIris/DeepStack, Home Assistant on VirtualBox, Plex Media Server, so it is very powerfull.
I also have a dual Google Coral TPU M.2 because initially i wanted to use Frigate but i changed my mind.
First question : is DeepStack can use Google Coral ? If not, Intel Movidius is a good choise ?
Second question : is there a way to set settings for DeepStack or i need to tweat it in BlueIris ?
Third question : DeepStack is not using my nVidia GPU but my i7 CPU or GPU i don’t know, what did I do wrong ?


I can just offer my suggestions for the second and third questions.
Second - As far as I know, DS is modified in BI. It has taken me quite a while to get set up, mostly because DS will only trigger when the BI motion detection is triggered. Then DS will analyze the image for the real time images X the analyze one each time. For instance, where real time images set to 5 images, and the analyze one each time set to 100 ms, when motion is triggered, DS will then try at time 0, +100ms, +200ms,+300ms,+400ms to see if the image fits one of the To confirm list items at the min confidence set. So setting the number of images and the analyze time is important based upon the object moving into view on the camera.
Third question. I first had DS CPU set up, then set up the GPU version. You have to completely delete the CPU version before installing the GPU version of DS on your PC. The cuda and drivers can be installed first, but then delete the CPU DS before installing the GPU version. As you have noticed, you should immediately see a CPU freetime difference when installed and working correctly. I have an I7 8700 with at GTX 1080, and saw CPU drop by 20-30% with 12 cameras, and GPU went up by about 3%.

This is what I have seen and my opinions. Hope they help.