GPU computing

GPU computing

GPU computing is the technique where GPUs are used for data processing replacing the conventional CPU. GPU’s parallel architecture has been very popular in the scientific community due to its capability to process massive amounts of data faster than a CPU.

GPUs were developed to render 2D graphics at the beginning but they are evolved to render 3D graphics. This fact gave GPUs enormous processing power. The focus of scientific projects is on their excellent floating point performance. This power is harnessed through GPU computing.

Generally speaking, a CPU has a few cores which are more proper for sequential data processing. On the other hand, GPU has a big amount (thousands) of little cores which make parallel processing very easy. The following schematic shows a comparison of how many cores exist in a CPU compared to a GPU:

A parallel way of CPU-GPU co-operation can be used to process a task: GPUS handles the intensive part while the rest of the code is executed by CPU:

The history of GPU computing is almost a decade. NVIDIA is the company which developed a C environment for executing parallel calculations on GPUs. This is the so-called CUDA environment. There are also other environments developed for GPU computing from other companies such as Apple and Microsoft.