Alphabet Inc’s Google has released new details about the supercomputers it uses to train its artificial intelligence models. Google has designed its own custom chip called the Tensor Processing Unit (TPU) which is faster and more power-efficient than comparable systems from Nvidia Corp. Google uses these chips for more than 90% of its work on artificial intelligence training, which involves feeding data through models to make them useful at tasks like responding to queries with human-like text or generating images.
Google has unveiled its fourth-generation TPU, a powerful AI supercomputer made up of more than 4,000 chips connected together with custom-developed optical switches. This new TPU is designed to power large language models, such as Google’s Bard and OpenAI’s ChatGPT, which are too large to store on a single chip. With its improved connections, the TPU is capable of processing large amounts of data quickly and efficiently, making it an ideal tool for AI research and development.
Google has developed a language model called PaLM, which is the largest publicly disclosed language model to date. It was trained by splitting it across two of Google’s 4,000-chip supercomputers over 50 days. Google’s supercomputers make it easy to reconfigure connections between chips on the fly, allowing for faster training and better performance. This new model is expected to revolutionize the way language models are trained, making it easier and faster to develop more accurate models.
The system uses circuit switching to route around failed components, allowing for flexibility and the ability to change the topology of the supercomputer interconnect to accelerate the performance of machine learning models. Startup Midjourney used the system to train its model, which generates fresh images after being fed a few words of text. Google’s supercomputer is a powerful tool for machine learning, providing a reliable and efficient way to train models and generate new images.
Google has announced that its fourth-generation Tensor Processing Unit (TPU) chip is up to 1.7 times faster and 1.9 times more power-efficient than a system based on Nvidia’s A100 chip. This is according to a paper published by Google, which compared the performance of the two chips. Google’s TPU chip is designed to help speed up machine learning tasks, such as image and speech recognition. Nvidia declined to comment on the comparison. The new TPU chip is expected to help Google’s cloud computing services become more efficient and cost-effective for customers.
The chip of this new TPU is not directly comparable to Nvidia’s current flagship H100 chip, as the H100 was released after Google’s chip and is made with newer technology. Google has hinted that they may be working on a new TPU that could compete with the H100, but they have not provided any details. Google’s TPU chips are designed to help speed up machine learning tasks, and the new chip could potentially provide even more powerful performance.
