I’m using CUDA on 32bit WinXP and I have the 8800 Ultra (768 MB Memory). The system has 2.5GB of RAM available. I’ve got the newest CUDA drivers and the newest CUDA SDK installed.
When I try to allocate more than 600MB in emulation mode the memory allocation fails. When allocating the memory on the device (Debug or Release build) everything works fine. I’m allocating with cudaMalloc(). Before allocating the System has about 1.5GB free RAM available, the application requires about 750MB RAM. Am I missing something here?