CodePudding user response:
This actually be divided into two parts, the first part is the industry trend, the other part is the need of hardware bottleneck, at present in the field of intelligent mobile phone chips, to implement the AI in one of two ways: the first is the traditional CPU + GPU + ASIC components to cooperate with other AI computing tasks, common graphics computing, face recognition, taking pictures such as the depth of field, the representative of the typical is qualcomm, it is to use the CPU + GPU + DSP to do AI task processing, but the AI application obviously can't just stay on the recognition level, such as video real-time beauty beauty, a higher level of graphics image pixel compensation, other applications such as human-computer interaction, GPU alone the (part-time) how line? So manufacturers began to adopt special - AI AI processor designed nuclear,