Annotation

Annotation: The article analyzes modern approaches and standards for describing general-purpose computing for graphics accelerators according to the GPGPU (General Purpose computing on Graphics Processing Unit) approach. The prospects for the implementation of an ultra-large integrated circuit oriented towards parallel computing in accordance with general-purpose tasks, while supporting the construction of the display of graphic images as an auxiliary function, are considered. Similar to the Nvidia Quadro GPU, such VLSI can be used for workstations that perform computational tasks such as CAD, solving systems of differential equations, spectral and wavelet analysis, statistical data processing, machine learning and similar classes of calculations running on a large number of parallel computing nodes. The presence of standard approaches to the description of general-purpose computing makes it possible to base the design process on existing high-level computing models, focusing on their hardware support. At the same time, performance in tasks that are typical only for the construction of threedimensional images can be reduced by increasing the efficiency of VLSI in general-purpose computing.