October 20, 2025 — Google Cloud has announced the general availability of its new G4 Virtual Machines (VMs), powered by NVIDIA’s RTX PRO 6000 Blackwell Server Edition GPUs. Designed to handle the next generation of multimodal AI, simulation, and graphics-intensive workloads, the G4 series delivers up to 9x the throughput of the previous G2 instances.

A New Standard for GPU Performance in the Cloud

The G4 VM family brings together NVIDIA’s cutting-edge GPU architecture with Google Cloud’s optimized infrastructure to power demanding workloads — from generative AI inference and fine-tuning to industrial digital twins and real-time 3D rendering.

Each G4 instance supports up to eight RTX PRO 6000 Blackwell GPUs and as much as 768 GB of GDDR7 memory, offering immense parallel compute power and memory bandwidth for AI, simulation, and visualization tasks. Fractional GPU options will be available soon, enabling customers to right-size resources for cost efficiency.

Key Features and Benefits

1. Generative AI and Multimodal Inference

G4’s FP4 precision support accelerates fine-tuning and inference for large language models (LLMs) and multimodal AI applications, including real-time text-to-image and text-to-video generation.

With Multi-Instance GPU (MIG) technology, each GPU can be partitioned into up to four isolated instances, optimizing price-performance and utilization for concurrent AI workloads.

2. NVIDIA Omniverse on Google Cloud

Google also announced NVIDIA Omniverse as a fully managed Virtual Machine Image (VMI) on the Google Cloud Marketplace. When paired with G4, developers can build and deploy industrial digital twins and AI-powered simulations using Omniverse libraries, OpenUSD, and NVIDIA Isaac Sim for robotics development.

3. Enhanced Multi-GPU Scalability

Thanks to a custom PCIe-based peer-to-peer (P2P) data path, G4 VMs achieve up to 168% higher throughput and 41% lower latency for tensor-parallel model serving — a major advantage for large-scale generative AI deployments.

4. Integration with Google Cloud Ecosystem

G4 VMs are fully integrated with:

• Google Kubernetes Engine (GKE) — for containerized GPU scaling with Autopilot and Inference Gateway

• Vertex AI — for seamless model training and deployment

• Dataproc — to accelerate large-scale Spark and Hadoop analytics

• Cloud Run — enabling real-time inference on a serverless GPU platform

Storage options include Hyperdisk ML for low-latency IO, Managed Lustre for high-performance parallel storage, and Cloud Storage for scalable dataset and artifact management.

What Industry Leaders Are Saying

“The combination of NVIDIA Omniverse on Google Cloud G4 VMs is the true engine for our creative transformation,” said Perry Nightingale, SVP Creative AI, WPP. “It allows us to generate photorealistic 3D advertising environments in hours, not weeks.”

“We’re excited to bring Google Cloud G4 VMs into Altair One, enabling demanding simulation workloads with the speed and visual fidelity to push innovation further,” added Yeshwant Mummaneni, Chief Engineer – Analytics, HPC, IoT & Digital Twin, Altair.

The Google Cloud Advantage

The G4 series extends Google Cloud’s NVIDIA GPU portfolio, sitting between the A-Series for scale and the G2 for cost efficiency. Combined with the new NVIDIA Omniverse offering, enterprises can now build, simulate, and deploy advanced AI and visualization workloads with lower latency, higher throughput, and simplified cloud scalability.

G4 capacity is available today in more Google Cloud regions than ever before. Developers can launch G4 VMs directly from the Google Cloud Console or deploy Omniverse and Isaac Sim via the Google Cloud Marketplace.