Which on-prem GPU server supports Custom Object Detection on Multi camera Stream using DeepStream 7.0

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, , ,
1

My Team is working on a project that involves processing video feeds from over 30 cameras simultaneously. We need to run multiple custom Deep Learning models on the data, and near real-time processing is a critical requirement.

Could anyone please advise on the most suitable on-prem GPU hardware server for this task? Specifically, we are looking for production grade use case where all the scenarios need to be considered, few mentioned below:

  1. Hardware Recommendations: Which NVIDIA GPUs or hardware servers would you suggest for real-time processing of video feeds from multiple cameras (30+) while running deep learning models?
  2. Architecture & Framework: What would be the best architecture and deep learning frameworks to support this use case? We are looking for high performance and scalability.
  3. NVIDIA Technical Support: Could you also provide details on the technical support offerings provided by NVIDIA for such use cases?

This request is on behalf of Accenture, and we would appreciate any official guidance or support you can offer.

Thank you in advance!

5

The Video Encode and Decode GPU Support Matrix | NVIDIA Developer and CUDA GPUs - Compute Capability | NVIDIA Developer may help you to get some data for the Nvidia GPU capabilities.

DeepStream SDK can be a candidate. Welcome to the DeepStream Documentation — DeepStream documentation 6.4 documentation

6

As per the GPU support matrix i can figure out that NVIDIA RTX A6000 or the NVIDIA A40 or P40 will be idle for the solution but wrt to “NVENC
/CHIP” number as 2-3 can the DeepStream concurrently process 20+ camera streams live using deepstream-test3 ?

If Yes, Can you share articles and repo links for DeepStream 7.0.

Also where can i find the link to the GPU to CUDA version mapping sheet.

7

For receiving and decoding RTSP streams, it is only related to hardware video decoder. The number of supported streams is decided by the stream format, resolution and framerate. There is performance data for 1080p@30fps encoding / decoding of different GPUs. Video Codec SDK | NVIDIA Developer

You may refer to How many H265 video streams can be decoded by a RTX A2000 12GB? - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums for how to explain the perf data.

For DeepStream 7.0. Please refer to DeepStream SDK - Get Started | NVIDIA Developer

The latest CUDA version supports most of Nvidia GPUs.

8

Is there any vendor tied with Nvidia to provide Embedded AI Edge Devices with preferred GPU inbuild for On-Prem processing.

If yes, Can you please share the link or process to procure it.