General Purpose Computing on Graphics Processing Units: From Fixed-Function Pipelines to Programmable Cores
Ehab Saleh, Chandrasekar Shastry · 2022
Nowadays, there is a huge demand for high-performance applications that require massive parallel operations that far exceed the capabilities of traditional CPUs. On the other hand, the rapid increase in performance and programmability of Graphics Processing Units (GPUs) has evolved these hardware units into the general-purpose computing field, placing the GPU at the heart of many high-performance applications. Despite the fact that programmers have access to thousands of general-purpose graphical processors, choosing the right device with specific features will play a significant role in the final decisions regarding the results obtained. Because NVIDIA has been a pioneer in changing the architecture of its GPUs to perform general-purpose tasks, we present all GPU architectures developed by NVIDIA Corporation over the last two decades in this paper. We will also look at the major hardware differences along with all the software platforms and solutions that have been introduced in these architectures. This survey should be of particular interest to researchers who are interested in using NVIDIA’s GPUs for developing their high-performance applications.