Google Cloud opens NVIDIA Vera Rubin A5X instances next month

Google Cloud and NVIDIA say that A5X instances, which are cloud servers, will open to all customers next month. Each instance uses NVIDIA Vera Rubin NVL72, a rack of AI chips, to train models and for agent work. Europe is next. The two companies give one test result, from WPP, but they give no price.

How it works

A5X has NVIDIA chips, network cards, the Google Cloud Virgo Network and liquid cooling in one system. It is for reinforcement learning, a method to train a model with tries of the model. Ineffable Intelligence is the first customer.

The numbers

WPP, a marketing company, says that work for a cluster of many servers operated on one A5X instance in its first tests. Google Cloud and NVIDIA gave no price and no speed numbers.

What is not known

Google Cloud also opens Confidential Computing on G4 virtual machines. It keeps models and data private on servers that the customer does not have. The two companies only test a method to increase performance for each watt. Use of energy is not known.

Insight: NVIDIA sells chips, protection and agent tools as one bundle

Google Cloud adds Confidential Bare Metal to A4X Max, and NVIDIA gave no speed numbers for it. WPP operated its agents on A5X with NVIDIA models and OpenShell, a sandbox tool. This reads as a bundle from NVIDIA of chips, data protection and agent software. A buyer of the chips can also use its rules. Leaders must find which parts they can replace.