I’m trying to get CUDA 8 and OpenCV to play nicely together, and I’m having some issues. I’ve posted in their forums as well, but I figured I’d cross post to see if anyone here had anything that might help. I’m trying to link CUDA8 and OpenCV statically to create some libraries and link them up the chain. When I run the OpenCV unit tests, any test using CUDA fails nearly instantly. After some troubleshooting, it looks like the allocation is where it’s dying, but I can’t see any further than that. This same setup with CUDA 7.5 works flawlessly. Under what circumstances will allocation on the GPU using cudaMallocPitch fail?