Google is throwing down the gauntlet in the machine-learning arena, flaunting the capabilities of its Tensor Processing Unit (TPU) silicon in a new paper. Claiming 15x the speed and 30x the performance per watt of Intel’s Xeons and Nvidia’s GPU, Google custom ASIC seems to clearly demonstrate the benefit of having exactly the right tool for the job, and likely the future of the market. Launched back in May, Google’s TPU is a chip designed to perform machine-learning functions using Google’s Tensorflow framework, and open source design that reached v1.0 back in February. Google used its TPUs in the AlphaGo competition, which pitted its AI system against the world’s best Go player – a game famed for being very hard…