I searched the forums and wiki but I could not find guidance on this question.
Use case: 12 cameras, main stream 4MP H.264 or H.265, 24x7 recording of main streams, modect (on sub streams, right?) to trigger machine learning object recognition. I will also probably create a summary video on 2-3 of the cameras once per day. I plan to use a Xeon E3-1270v2 as the CPU, please let me know if that is under powered.
VRAM:
I see in the docs that there is mention of a 1080 ti used by the developer that has 4GB VRAM and there was a need to modify the model because it consumed too much VRAM. Is this impacted by the model, the number of cameras, resolution of images, or other factors? Would a card with 2GB VRAM be sufficient? (750 ti) Should I use a more modern card with more VRAM (1650 with 4GB)? Most helpful would be guidance on how to calculate the needed VRAM myself.
GPU:
The examples in the docs are using a 1080 ti and this is still a pricey card (to me, for a NVR). Does anyone have data on cheaper cards? I have a spare 750 ti but have also considered a 710 or 730 if they will be faster than the CPU.