One of the reasons why Apple implemented OpenCL technology in Snow Leopard to meet the growing demand for computing power in scientific applications, in areas such as mathematics, engineering, astronomy, chemistry and biology. Using the GPU of modern computers as a second processor, we already know that it is possible to achieve huge gains in processing power in the execution of any application, benefiting all users.
However, in more demanding applications for scientific purposes, researchers have found very high performance gains when using modern GPUs for general computational purposes. An article recently produced by MacResearch.org it proved this by presenting the results of a simulation carried out with astronomy software, aimed at visualizing physical effects with graphic modeling in real time.
The system used during the tests was an IBM Blade QS22 with 16GB of RAM, AMD Phenom processor quad-core 2.5GHz and two more 3.2GHz Cell BE processors each, in addition to an NVIDIA Tesla C1060 GPU. With this hardware, a particular application compiled to take advantage of the two Cell chips was 25 times faster in presenting graphical results when compared to the CPU processing everything by itself. The same software was compiled to work with the Tesla GPU and was 4% faster than Cell chips while performing the same task.
Such a performance increase has rarely been seen when working with applications that are highly optimized to take advantage of certain types of hardware, but it doesn't stop: when compared to proprietary technologies for parallel computing (such as NVIDIA's CUDA), OpenCL is also a little faster. This is particularly interesting for companies and research centers, especially considering that OpenCL can be used not only on Mac OS X Snow Leopard, but also on Linux systems (as was the case with the above test) and Windows.