Running on a multi GPU server can not find an example of DeepStream-Triton config file and source detection set up file that make use of more than 1 GPU (gpu-id=0). Is there an NVIDIA internal one that you might have and if so could you please post it?
Am looking to have each GPU take a slice of the incoming frame stream and perform object detection. A RetinaNet/ ResNet example would be great!
**• Hardware Platform (Jetson / GPU) Mult-GPU
• DeepStream Version DS 6.0
• JetPack Version (valid for Jetson only)
• TensorRT Version
**• NVIDIA GPU Driver Version (valid for GPU only) 11.7
Hi @brandt33 ,
Is it possible to share a pipeline diagram you expect?
Hello @mchi, Thank you for asking, here is an idea of what I was thinking about …
DeepStream-Triton Example.pptx (41.3 KB)
Thanks for your slide!
if each GPU takes a slice, there is a metamux needed on the output, but crrent DS does not support this.
We are actually working a similar sample for such use case: running multiple models in parallel on multiple branches on selected GPUs, user can select different streams for different branches, and mux the meta for the same frame at last. This sample will be public in near future,
Thank you @mchi for the information, that is great to know and also helps me understand what the current possibilities are!