How to cuda11 install and configure Caffe in deepstream Python API？

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1
**• Hardware Platform (Jetson / GPU)：dGPU**
**• DeepStream Version:5.1**
**• JetPack Version (valid for Jetson only)**
**• TensorRT Version:7.2.3**
**• NVIDIA GPU Driver Version (valid for GPU only):11.1**
**• Issue Type( questions, new requirements, bugs):questions**

Hello, teacher, how can cuda11 configure Caffe to model in the deepstream Python API? Because others say that Caffe was not updated three years ago and does not support cuda11. What should I do?

3

You can choose to use onnx model

4

Hi, friend, thank you for your answer. I didn’t know that the onnx model could also be used. Can you teach me how to change it? Please, I really need your help

c8873c93574e1c82f584d541dff9b54
c8873c93574e1c82f584d541dff9b541641×826 72.4 KB

5

If you are using the DS built-in models. then it should work out of box, please follow the readme to setup

6

image
image1344×859 50.9 KB

Do you mean that I can only provide models in the corresponding format according to his format？

7

It is said that NVIDIA has a Caffe version that is independently maintained and updated internally. How can I use it to transfer my model?

8

Yeah, the model inside DS package is caffe model, it should work with cuda11.

9

image
image960×330 9.9 KB

My current graphics card is rtx3080. Officials say that the current Caffe does not support 30 series graphics cards. I tried to install it, but it really didn’t work. What should I do?

10

Can you share the link with us?

11

OK, wait a minute

12

For example, here, I try to change the CUDA version to cuda11.0 corresponding to RTX3080, and the cafe compilation will fail

13

Thanks, we need to confirm the following things before we do further check on deepstream side

  1. I’m not sure whether the blog you referred is official, seems it’s a personal blog, is there any official document or announcements from Nvidia or maybe the caffe organization.
  2. Also if that’s true, then you need to create a topic on cuda forum since it’s not a DS related issue.
14

OK, first, you mean NVIDIA has its own cafe, but I haven’t found a way to install it, have I?

15

Because all Caffe installation methods I can find are personal blogs. If you know that there is an official installation method, please send me a connection. Thank you

16

No, also I think TensorRT shuold consume the caffe model, seems you don’t need to install other caffe components.
Better to create a topic on cuda or TensorRT forum to get better explaination.