I need a cost calculation to support TensorFlow users on the cloud,I'm meaning about the predication(inherence) calculation cost to be done in the cloud and not in the end edge(user).(Need to support TensorFlow pose estimation libraries only)
For example, server cost and system required to support 10K users that need prediction(inherence) using TensorFlow pose estimation libraries.
More detailsRequests are distributed in time uniformly.
What are the hardware requirements? This is exactly the question that I'm looking to answer,I search and this question has different answers, some places require for pose estimation tracking(prediction only) 4 CPU and 1 GPU, but there exist different solutions like openvino that do not require GPU or TensorFlow lite that can run on android devices with less.
How long take only one prediction? There are different metric. 1 runtime prediction can take: ~100msAnother metric: 26 FPS.Do you have latency constraint? I assume 700 ms it's good enoughDo you have multi-region constraint? No.just USA