Comparison of architectural differences between CPU, GPU, and NPU

CPU/GPU/NPU Architecture Features

A picture is worth a thousand words. Regarding the structural differences between CPU, GPU, and NPU, as shown in the following figure:

From the perspective of optimization, the objective function of CPU is "minimize latency", GPU is "maximize throughput", and NPU further improves the utilization rate of storage on the basis of GPU.

Let’s start with the difference between CPU and GPU. The CPU is a central processing unit, which is a logical unit. The CPU is responsible for complex and small serial calculations. The GPU is mainly responsible for a large number of simple parallel calculations, as shown in the following figure:

Both GPU and CPU belong to general-purpose computing power, and the difference from NPU is obvious. Let’s compare GPU and CPU first, and finally discuss NPU. Functionally, GPU cannot replace CPU. Similarly, CPU cannot replace GPU.

Essentially, the difference between CPU and GPU is that the CPU is a vector machine & scalar machine, and the GPU is an array machine.

Guess you like

Origin blog.csdn.net/tugouxp/article/details/125719383