Mobo Memory as swap or main memory for GPU

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi

Is it possible to use motherboards memory as swap memory for GPU computation. If i would be dealing with large matrices that would exceed gpu memory, it would be impossible to copy over it.
So in those cases, is there a solution to work around that?

2

You can use motherboard memory directly by mapping it into the GPU’s address space (see section 3.2.4.3 of the Programming Guide). Probably it is more efficient though to split the matrix into tiles (submatrices) and use streams to parallelize operating on one tile with copying in the next.