Google Claims Its New TPUs Are 1.7 Times More Powerful Than Nvidia’s A100 Chip, Which Is Used By ChatGPT

Google has revealed that its latest 4th generation tensor processing units (TPUs) are 1.7 times faster than Nvidia's A100 chip. The A100 is the basis of the Microsoft Azure cloud compute architecture, which was built to run OpenAI’s GPT-4 and ChatGPT. Google also claims that its processors are 1.9 times more power efficient than Nvidia's chip, making them more energy-efficient.

It is important to note that Google is not comparing its TPUs to Nvidia's new H100 chip, which is also used to power GPT-4 and new Microsoft applications like the new Bing search engine that harnesses the power of OpenAI’s generative AI assistant.

Google released a scientific paper revealing this information. In fact, this 4th generation TPU system has been active in Google's data centres since 2020. Google also mentioned that midjourney, a generative AI startup, used them for training their models to create new images with realistic graphics after being fed a few words of text.

Google has hinted that it is working on a new TPU that will be more comparable to Nvidia's H100.

More than 90% of Google's AI development work happens on TPUs. In its 4th generation chip, Google strung 4,000 of the chips together to create a supercomputer. It also created custom optical switches to help connect the individual supercomputers.