At Google IO conference, the company made a small splash. The company admitted that over the years developed a special processor for artificial intelligence. Furthermore, processors called Tensor Processing Units (TPU) for over a year serving on the company's data centers.
According to Google, the processor (or accelerator) Tensor Processing Units is three generations ahead of competitors like development. It is designed to speed up the tasks associated with deep machine learning and much more effective than the use of general purpose processors or accelerators based on FPGA-matrices. The company did not share the specifications of new items, and hinted that the TPU processors do not go beyond the limits of computing Google halls.
It should be noted that the TPU module is quite original. It has dimensions similar to the dimensions of a typical hard disk and installed in the same interface connector as HDD-drive. In fact, Google uses to create the AI counters conventional baskets for hard drives. It seems that this has helped the company to keep secret development. By the way, the source code TensorFlow system to promote learning machine company made ??public in the last year . A processor for optimum performance code revealed only now. Related Products :
|