Installer cuda check

Software
1

The ZED SDK 4.0.2 installer is installing it’s own version of cuda despite the base for our docker image ( nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04 from Docker) already having it. If I run with skip_cuda then ZED_Diagnostic complains that cuda cannot be detected. This second installation severely bloats our docker image to an infeasible size.

Can you explain what the installer is checking for and how I can have it recognize the base versions cuda. I’ve been reading through your image layers from stereolabs/zed:3.8-runtime-cuda11.7-ubuntu20.04 from Docker , however its not clear what is required.

Also do you plan to updated zed-docker/Dockerfile at master · stereolabs/zed-docker · GitHub

2

If we ignore ZED_Diagnostic output for now, can you successfully run ZED SDK samples with the default cuda installation (ie using skip_cuda)? If not, are standard CUDA samples working?

ZED_Diagnostic may not be relevant in some cases, especially in docker.

We’ll update the zed docker repository for the 4.0 GA (stable) release