CUDA OpenGL interoperability problem (Texture memory)

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello!
I’v encountered the following issue with OpenGL CUDA interoperability: when I create a texture with glGenTexture, glTexImage functions and then register it with cudaGraphicsGLRegisterImage I’m getting double memory consumption - i.e. if texture data was 256MB I’m getting around 512MB consumed…
Is it normal behavior? I hope it is not…
Can somebody explain what is happening?

2

So, have anyone accountered the same problem? Or the bug just have no solution?

3

I’ve run into the same exact issue with Windows 10 with both Maxwell and Pascal architectures. However, it is not an issue on Kepler and Maxwell architectures with Windows 7. I’ve updated to the latest versions of CUDA and NVIDIA drivers to no avail. We’re you able to solve this issue?