From the point of view of the hardware architecture of the server, AI servers are servers in the form of heterogeneous, and different combinations can be used in the heterogeneous way according to the scope of the application, such as CPU+GPU, CPU+TPU, CPU+other acceleration cards and so on. Compared with ordinary servers, there is no difference in memory, storage, network, mainly in the big data and cloud computing, artificial intelligence and other aspects of the need for greater internal and external storage, to meet a variety of data collection and organization.
We all know that the ordinary server is a CPU as the provider of computing power, the use of serial architecture, in the logic of computing, floating-point type of computing and so on is very good at. This is because a lot of branch-hopping processing is required to make logical judgments, making the CPU structure complex, and the increase in computing power is mainly achieved by stacking more cores.
But in the application of big data, cloud computing, artificial intelligence and the Internet of Things and other network technologies, the data flooding the Internet has shown a geometric growth, which puts a serious test on the traditional services with the CPU as the main source of arithmetic, and in the current CPU process technology, the number of cores of a single CPU is close to the limit, but the increase in data continues, so it is necessary to improve the server's data processing capacity. Therefore, in this environment, AI servers were born.