(1) Performance is very different from traditional chips, such as CPU and GPU. Faster and more energy efficient when executing AI algorithms.
(2) There is no difference in process, everyone is the same. At least for now, it's all the same.
< br >
The so-called AI chip is generally a pointer to the ASIC (special chip) of the AI algorithm.
Traditional cpus and Gpus can be used to execute AI algorithms, but the speed is slow and the performance is low, and it cannot be actually commercialized.
For example, automatic driving needs to identify road pedestrian traffic lights and other conditions, but if it is the current CPU to calculate, then it is estimated that the car turned into the river and did not find the river in front of it, which is slow, time is life. If the GPU is used, it is indeed much faster, but the power consumption is large, the car's battery is estimated to be unable to support normal use for a long time, and the old Huang's GPU is extremely expensive, often tens of thousands of single pieces, ordinary consumers can not afford to use it, and often out of stock. In addition, because the GPU is not an ASIC developed specifically for AI algorithms, in the final analysis, the speed has not reached the limit, and there is room for improvement. And in areas like intelligent driving, you have to be fast! In the mobile phone terminal, you can own face recognition, voice recognition and other AI applications, which must be low power consumption, so GPU OUT!
First, the performance is very different from traditional chips, such as CPU and GPU. Faster and more energy efficient when executing AI algorithms. Artificial intelligence The algorithm of the AI chip is more advantageous
Second, the process is no difference, everyone is the same. At least for now, it's all the same. The NPU unit of the artificial intelligence AI chip is more powerful
,,,,
The biggest difference is the difference between mobile and server, and there are many people who call the two categories terminal and cloud. Artificial intelligence is more terminal, and AI chips are more cloud.
In fact, in these two types of applications, artificial intelligence chips have essential differences in design requirements. For example, the mobile side pays more attention to the low power consumption, low latency, and low cost of AI chips, while AI chips deployed in the cloud may pay more attention to computing power, expansion capabilities, and its compatibility with existing infrastructure, and so on.
For these two types of AI chips, it is difficult for us to directly compare. It's like a big tree, its trunk is responsible for supporting the tree and also transporting various nutrients. Its leaves are responsible for photosynthesis and the production of nutrients. But it is difficult to compare the trunk with the leaves, which is more useful.