With the new CUDA 6.0 Release Candidate now available for download, CUDA developers can start saying goodbye to cudaMalloc. Since the first CUDA SDK was released almost exactly seven years ago, one of the first things every developer who learned CUDA mastered was the concept of different system memory and GPU memory and using cudaMalloc to allocate GPU memory. Now for the first time with CUDA 6, developers can have a new unified view of memory.
The new CUDA 6 unified memory is created with a new call, cudaMallocManaged, making it super simple to get started developing CUDA code.
For those of you just getting started with CUDA, there are a great set of tutorials and other resources located on the NVIDIA Developer site.