when I use cudaMalloc() to malloc gpu memory, I find host memory (physical memory and virtual memory) is increased in System commit of Windows task Manager. The increased host memory size is a little larger than the gpu memory size allocated by cudaMalloc(). I want to know how and why it happens. If this is rational,should I keep 1G host memory(physical memory and virtual memory) to allocate 1G gpu memory by cudaMalloc().
Microsoft Windows 7 64bit
GeForce GTX 560 Ti
Driver Version:311.06(NVIDIA CUDA5.0.1)