Not long ago, Google go robot Alphago swept the Korean king Li Shishi, the data show that the AI (artificial intelligence) equipment piled up a lot of top chip.
With a minimum of 48 CPUs and 1 GPUs, the highest configuration, Alphago, is known as Alphago distributed, which was installed in March as 1920 CPUs and 280 GPUs, and is constantly being upgraded. At that time, the external chip providers were Intel and Nvidia. However, from the I/O conference on the new news, Google revealed that the defeat of South Korea's set of Alphago in fact built in their own design of the microprocessor.
It is reported that this microprocessor chip belongs to the application ASIC (ASIC), has been secretly developed for three years, the design of the first deployment of artificial intelligence, deep neural technology. In Google's documentation, the name of this processor is TPU (strong processing unit).
After three years of secret research and development, Google said, TPU leading the current machine learning processor (CPU, GPU) more than 7 years, that is, three generations, 10 times times faster.
Google has abandoned Intel, NVIDIA's secret development processor.
Although TPU can not directly replace the common CPU, GPU in our life, but in the future of the forefront of the AI field, Google will be a lot less reliance on intel\amd\nvidia, especially in the field of the most aggressive, the most product-rich Nvidia.
This address: http://www.linuxprobe.com/google-dev-processor.html
Google abandoned Intel, nvidia secret development processor