Question about cudaGraphicsMapResources speed.

I’m writing to a texture in Direct3d 11 then reading from it in CUDA. As suggested in documentation and examples, I’m using cudaGraphicsMapResources on on every frame, and sometimes cudaGraphicsMapResources takes rather long time. It may take 10-30 ms in certain cases. It causes a framedrop in the application. The texture is only written in directx and only read in CUDA, not used anywhere else.

What can cause such a slow speed?

I have GeForce GTX 670, Windows 7 64 bit, 327.23 drivers.