Google Cloud Launches Two Eighth Generation TPU AI Chips

Google Cloud has introduced two new Artificial Intelligence (AI) chips for the TPU line, namely TPU 8t and TPU 8i. This 8th generation TPU is specifically designed for the development of AI agents.

"We are proud to announce our eighth generation of TPU, which is specifically engineered for the agent era," Google Cloud said on its official website, quoted on Thursday, April 23. According to him, these two new chips can support the AI ecosystem in a more interactive and cost-effective way.

TPU 8t was developed as the main engine for the training process of AI models with very high throughput. This device offers computing performance up to three times faster than the previous generation to shorten the training time of large-scale models.

One TPU 8t superpod unit can accommodate 9,600 chips that are integrated through high-speed interconnects. With the power of more than a million chips in a cluster, the training process of complex models that usually takes months can be completed in a matter of weeks.

Meanwhile, TPU 8i is developed as a special reasoning system for inference processes with very low latency. This chip is equipped with a high bandwidth memory of 288 GB to overcome data access barriers in dense AI workflows.

TPU 8i also includes a doubling of SRAM memory capacity on silicon to accommodate massive cache data. This design is believed to provide 80 percent better performance than the previous generation.

These two latest chips will be available to Google Cloud customers in the near future. With the arrival of TPU 8t and TPU 8i, developers now have greater flexibility in choosing hardware that matches the development stage of their AI.