NVIDIA and Google Cloud partner to offer 960,000 Vera Rubin-powered bare metal servers.



NVIDIA has announced it is expanding Google Cloud AI Hypercomputer for AI factories that drive next-generation agent-based and physical AI. Enterprises will now have access to 'A5X' instances powered by NVIDIA Vera Rubin, which can scale to up to 960,000 Rubin GPUs.

NVIDIA and Google Cloud Collaborate to Advance Agentic and Physical AI | NVIDIA Blog
https://blogs.nvidia.com/blog/google-cloud-agentic-physical-ai-factories/




The newly announced 'A5X' is a bare metal instance that utilizes the NVIDIA Vera Rubin NVL72 rack-scale system. A bare metal instance refers to a system in which a single physical server, rather than a virtual server, can be exclusively used by one company.

A5X reportedly reduces inference costs per token by up to one-tenth compared to the previous generation, and improves token throughput per megawatt by up to ten times. By combining the networking system 'NVIDIA ConnectX-9 SuperNIC' with 'Google Virgo,' it can scale to up to 80,000 NVIDIA Rubin GPUs in a single-site cluster and up to 960,000 in a multi-site cluster.

In addition, various systems can be integrated, including a preview of Google Gemini on Google Distributed Cloud running on NVIDIA Blackwell and NVIDIA Blackwell Ultra GPUs, confidential VMs with NVIDIA Blackwell GPUs, and agent-based AI on the Gemini Enterprise Agent Platform using the NVIDIA Nemotron open model and NVIDIA NeMo framework.

NVIDIA announces 'Nemotron 3 Super,' a hybrid MoE openweight AI model with 120 billion parameters and Japanese language support - GIGAZINE



NVIDIA emphasized that these services enable optimization of all workloads, from Mixture-of-Experts inference and multimodal inference to data processing and complex simulations of physical AI and robotics.

The partnership between Google and NVIDIA is reportedly reaching its 10th anniversary.

Related Posts:

in Hardware, Posted by log1p_kr