Hardware recommendation real time computer vision on many streams

Hi, we want to analyse more than 300 simultaneous streams of video real time on 640 x 480 10 fps. For detecting Person, cars and motors. Ee have a budget of 20K.

Do you think this is possible, with the budget? if not, what do you think is the most amount of streams we can analyse.

For that what think is the best hardware for doing this.

Thanks!

Basically, you need to check

  1. if the device support to decode so many fps at the resolution
  2. if the model you want to deploy on the device support the fps at the resolution

fyi - https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Performance.html

Thanks!

How i can check this?

  1. if the device support to decode so many fps at the resolution

Other question is linear the relation between fps and number of fps.
Example: 1T4 1000fps 2*T4 2000 fps

it seriously depends on the model you want to run, e.g. https://developer.nvidia.com/deep-learning-performance-training-inference

You can use TensorRT trtexec to benchmark your model , fyi - https://developer.nvidia.com/blog/speed-up-inference-tensorrt/, and https://github.com/NVIDIA/TensorRT/blob/master/samples/opensource/trtexec/README.md