by Anton Shilov
08/11/2005 | 02:55 PM
Graphics chip designer NVIDIA is describing opportunities for general purpose computations on graphics hardware, something which has a prerogative of enthusiasts for years.
<%BANNER[article]%>One of the courses NVIDIA hosted at Siggraph conference presented a detailed introduction to general-purpose computation on graphics hardware (GPGPU), with emphasis on core computational building blocks ranging from linear algebra to database queries. The course was held by NVIDIA’s engineers, which may indicate that computations beyond computer graphics on graphics chips are explored at NVIDIA.
Modern graphics processing units (GPU) are far more complex than contemporary central processing units (CPUs) and even though they work on much lower frequencies, the number of operations they can perform at once may exceed than of the CPUs. Unfortunately, in order to make GPU calculate something other than graphics, a completely new programming language, or a special translating software, should be written.
Several groups of enthusiasts have explored possibilities to use graphics processing units for general purpose computing, however, no project has actually become a success.