GPU technology trends and future requirements

Inherent massive parallelism makes GPUs (Graphic Processing Units) a natural beneficiary of Moore's law (1). Even modern GPU chips that contain billions of transistors are a long way from achieving perfect simulation and rendering of real-world scenes. The desire to have more transistors continues. In addition, the use of GPUs for parallel computing is only beginning to enable many exciting applications. This paper begins with GPU basics and its evolution with technology, then connects GPU requirements such as FLOPS, power and product variations to device and process technology. It will also address technological opportunities and challenges for future GPU advancement.