I’ve just installed the newest drivers and toolkit for CUDA 3.0 (197.13) but I’ve almost immediately run to quite a serious problem. In my application I’m mapping few huge 3D textures from openGL to CUDA which worked perfectly using the older CUDA 3.0 BETA drivers ( 195.39 ) but with the newest drivers whenever I call “cudaGraphicsGLRegisterImage()” CUDA allocates new device memory for the texture data. So for example in my case when I try to register a 200MB 3D texture, CUDA allocates another 200MB of device memory when I call cudaGraphicsGLRegisterImage().
Is this the expected behavior? I hope not. The strange thing is that it worked just fine with the older BETA drivers and I can still map all these textures into OpenCL without any problems even with the newest drivers which is kind of weird since OpenCL should be built on top of CUDA.
my OS is Win7-64