Google reportedly adds two TPU variants for secure and long-context AI

What happened

Google has introduced two new versions of its eighth-generation tensor processing units, called the TPU 8tx and TPU 8i Max. The TPU 8tx is designed for secure and confidential workloads, while the TPU 8i Max is intended for AI tasks that require processing large amounts of data at once. These new chips are updated versions of existing models and expand Google's TPU lineup to four specialized chips. Google rents these chips through its cloud service rather than selling them directly.

Key details

• TPU 8tx is aimed at secure, confidential work for government and regulated industries.
• TPU 8i Max is built for AI applications needing long context windows, handling large data inputs.
• Both chips are based on the eighth-generation TPU 8t and TPU 8i, first shown in April 2024.
• Google collaborates with AMD for the 8tx and works with four design partners, including Broadcom and MediaTek.
• The chips will be generally available later in 2024.

Why it matters

Cloud customers, including government agencies and companies in regulated sectors, now have access to a TPU specifically designed for secure workloads. Organizations developing AI models that require processing large amounts of data in a single request can use the new 8i Max chip. Google’s cloud clients have more specialized hardware options, which may help them address security requirements and performance needs for advanced AI applications.