To be fair, he did give a definition at 0:27, starting by saying it is a "specialized processor". I also would have liked to have seen a chart or table of the NPUs from different companies.

@@PunmasterSTP Yes. But I know what a NPU is, what he says there doesn't cut it. Filling a video about a tech topic with nice visuals is misleading when people want to learn anything

@@ThomasGodartI think this video was just intended to be a very basic introduction for lay audiences. Is there another video or channel you've found that you like better?

Although both TPUs and NPUs are designed to work specifically with deep learning models, so they may seem like the same type of processor, there are some differences between them: Purpose: TPUs are designed to accelerate the processing of tensor operations, used in image processing and convolution-based deep learning. NPUs, on the other hand, are specifically designed to accelerate the processing of deep neural networks. Architecture: TPUs are designed as a System on a Chip (SoC) that contains a large number of parallel processing cores, while NPUs are generally built as a separate processor or as part of an SoC. This means that TPUs often have a greater number of processing cores than NPUs. Performance: TPUs are known for being highly energy-efficient and offer exceptional performance for specific applications, such as deep learning. NPUs, on the other hand, may be less energy-efficient but are very versatile and can be used in a variety of artificial intelligence applications.

Apu