Among these network service companies, the most active is Google, whose self-developed chip TPU (Tensor Processing Units) has been developed to the second generation. Compared with the GPUs currently available on the market, Google's chip can machine learning. Training time is reduced by half.
As mobile chip maker Qualcomm intends to abandon its server division in May this year to focus resources on the mobile chip business, the industry believes that Microsoft's pace of self-developed chips will accelerate, because Microsoft reportedly has been out of the R & D engineers from the Qualcomm server division. .
Earlier, Microsoft said that its second-generation MR (Mixed Reality) helmet HoloLens will use the company's self-developed AI chip HPI (Holographic Processing Unit) to process the data collected by the sensors installed on the device, including depth. Sensing, head tracking and inertial measurements, etc. The HPU chip will also be used for other hardware devices from Microsoft and will also be licensed to other vendors for use on their terminal devices.
According to industry sources, the diversification of artificial intelligence (AI) applications requires the support of different chip solutions. In addition, Intel has long dominated the development and supply of processor chips, which prompted the network service giant to independently develop the chips needed. To spread the risk of over-reliance on Intel chips. Facebook, Amazon and Alibaba will soon start the chip development.
It is expected that in the processor field of many terminal devices, Intel's future performance will be dropped. The source said that the launch time of Intel's 10 nanometer manufacturing process Ice Lake processor platform will be postponed from the original planned 2019. Intel is currently facing serious challenges from AMD's Rome and Milan processors. AMD's two chips are produced using TSMC's 7-nanometer manufacturing process.