By Yina Moe-Lange
On Wednesday, Google announced its second-generation Tensor Processing Units (TPUs) Machine Learning chips. These updated chips are designed to be faster and more capable than at Machine Learning tasks than the first-generation chips, CPUs and GPUs. Google has designed them specifically to be used with Google’s TensorFlow framework for AI workloads. Also significant is the fact that Google is making these chips available as a service on the Google Cloud Platform (GCP).
Google’s announcement comes only a week after Nvidia announced their new Volta V100 processor. This GPU chip is part of Nvidia’s strategy to get Machine Learning embedded across all markets, specifically focusing on developing neural net programming – also know as Deep Learning. Both Google’s and Nvidia’s new chips are ramping up the processing speeds, vital in Machine Learning models.
Nvidia’s stock has skyrocketed over the past year as they became the frontrunner in Machine Learning hardware. The Machine Learning and AI boom has been heavily dependent on the speed and capabilities of the Nvidia chips and as their stock price reflects, many have assumed they are supposed to run the Machine Learning hardware in the future. With their newly announced chip, Google is on a collision course with Nvidia in this battle to create the processing chip of the machine learning future.
Two big questions remain to be answered. First, will the TPU have any significant impact? The current TPUs are only for inference and the new second-generation chips are supposedly built to be able to used for training Machine Learning models. If the second-generation TPUs fail to excel at training models, then in the mean time CPUs and GPUs will reign supreme. But, if they are successful in being able to drastically reduce training time, then the chip domain will begin to alter.
Secondly, if Google’s TPUs are successful, what will happen to Nvidia? As more tech giants begin to build their own hardware, will Nvidia’s position as the leading AI and Machine Learning chip maker be in jeopardy?