A typical AI server is built around a central processing unit, memory chips, and a processor designed for fast calculations. In all these areas, there have been many innovations, leading to faster speeds and more computation power. Also, because servers can work together to perform a joint analysis of the extremely large data sets. Notable developments in this field are the Nvidia and, more recently, AMD accelerator cards (GPUs), which include extremely fast processors. These have provided the necessary computer power to be able to calculate the latest AI models. A rule of thumb for technological innovation in the semiconductor industry is Moore’s Law, which states that the number of transistors on a microchip (providing computing power) doubles every two years. We therefore expect that further advances in computing speed are on the horizon, enabling more complex models. More efficient microchips could also lower the power consumption of the existing processing power: promising developments. - ING
Opmerkingen