Different texture lookup value emudebug and debug where to start looking?

Accelerated Computing CUDA CUDA Programming and Performance
1

Can anyone offer some insight as to where to start looking for a bug for the following case:

Setting up a 2d texture of floats using cudaMallocArray and cudaBindTextureToArray. The textures are not power of 2’s but that shouldnt be a problem (?).
width: 20 000, height: 20.
Using tex2D within the kernel.
Linear interpolation is disabled

Emudebug returns the correct values
Running the code in debug mode and writing back the lookedup values to the host results in the wrong value (and not off by one… off by many!)

Thanks!