What does GP-GPU computing mean?
Quite simply, it is the combining of the best attributes of the CPU and the GP-GPU so that each processor works collaboratively to maximize the performance of massively parallel applications. Nvidia, with its CUDA programming toolkit, works to allow serial portions of code to execute on the CPU, and parallel portions to execute on the GPU.
Another way to look at is to say that heterogeneous CPU + GPU co-processing is the using the right core for the right job. We use a CPU core (optimized for low latency on a single thread) for a code’s serial portions, and we use GPU cores (optimized for aggregate throughput on a code’s parallel portions) for parallel portions of code. This approach gives more performance per unit area or power than either CPU or GPU cores alone.
