Help decide Nvidia Hardware based on number of video streams it can process

I need to decide which GPU hardware will better suit the requirement, below is my question and I would need some help choosing the right hardware

  • How many RTSP video streams can RTX 3050 & 3030 can process based for PeopleNet-ResNet34 BodyPose3D Model

  • Can Jetson AGX Orion better than using regular GPU enabled server but Graphics card added server seems to be better value for the number of cuda cores 10 times higher than Jetson devices for the similar cost?

  • Power usage constraint is not at all a problem for this requirement just cost/video stream matters

Im not getting there is always benchmarks comparing Jetson devices more than Geforce hardware, in a typical video analytics solution power is not a constraint and what else benefit we get when using Jetson over Geforce???

Hi @kbalu,

Number of videos stream that can be supported on a particular GPU will depend on the use case, tech stack used and kind of pipeline optimization that has been done on the end-end solution.

But you can refer to below links to understand the FPS results achieved on various GPU for a particular scenario:


There isn’t sufficient information that clearly guides us to buy hardware and decode Nvidia marketing of Jetson vs RTX for inference use case (Its a simple question, we don’t want accurate numbers) and we don’t plan to train the model here

Just simple as this, what Nvidia recommends for video inference application, RTX or Jetson, Nvidia seem to push Jetson when power consumption is a not at all concern for MOST real time video analytics solution except robotics/mobile devices, as these are usually deployed to on premises

Jatson family or RTX family for inferencing for similar cost of device?

I would not compare Jetson vs a PCIe based GPU (RTX, A100, etc) solely based on inferencing. You are talking about an embedded system primarily targeted for edge and far-edge applications when close proximity to the sensors is a factor. A PCIe based GPU can be deployed from the edge to the cloud but needs to be integrated into a larger system.

We recommend both systems for video inferencing. Both product lines have different options at different price points. It depends on your overall application requirements (performance, deployment, cost, power, etc).

However, if you solely want to look at cost, I guess you can say that, assuming either device can meet your design needs then the Jetson might have an advantage solely based on the fact that you would save on the cost of the system needed to use a PCIe based GPU.

However, if you are designing a system targetted for production I would not limit your analysis to a single factor.