In today's Google I / O Conference, artificial intelligence accounted for a large proportion, covering the software, hardware and services three large pieces of content.
Among them, the hardware part of the most dazzling is Google second generation TPU processor.
According to reports, TPU is Google independently developed a conversion to AI computing services, high-performance processors, the first generation of products in the data center and including AlphaGo, such as artificial intelligence which use, the main operational performance. And today released the second generation of TPU is to help deepen the artificial intelligence in learning and reasoning capabilities, the current Google has already started to use.
Google said,The second generation TPU can provide amazing 180TFLOPs floating point computing power, compared to the traditional GPU and CPU were 15 times higher and 30 times.
At the same time, Google also supports an operator array called TPU pod,Each TPU pod includes 64 second-generation TPUs that provide 11.5PFLOPS floating-point capability.
In addition, in order to allow artificial intelligence developers to better enjoy the performance bonus of the TPU, Google in the I / O Conference also officially launched the Cloud TPU cloud computing services, Google cloud services can be rented in the future. As for the charges will also be taken and the original GPU computing services the same charging model, measured in minutes.
Second generation TPU unit