GPU-Accelerated Google Cloud Platform

Combine the power of the NVIDIA AI platform with the flexibility and scalability of GKE to efficiently manage and scale generative AI training and inference and other compute-intensive workloads. GKE's on-demand provisioning, automated scaling, NVIDIA Multi-Instance GPU (MIG) support, and GPU time-sharing capabilities ensure optimal resource utilization. This minimizes operational costs while delivering the necessary computational power for demanding AI workloads.

Combine the power of NVIDIA accelerated computing with Google Cloud’s Vertex AI, a fully managed, unified MLOps platform for building, deploying, and scaling AI models in production. Leverage the latest NVIDIA GPUs and NVIDIA AI software, like Triton™ Inference Server, within Vertex AI Training, Prediction, Pipelines, and Notebooks to accelerate generative AI development and deployment without the complexities of infrastructure management.

Leverage the NVIDIA RAPIDS™ Accelerator for Spark to accelerate Apache Spark and Dask workloads on Dataproc, Google Cloud’s fully managed data processing service—without code changes. This enables faster data processing, extract, transform, and load (ETL) operations, and machine learning pipelines while substantially lowering infrastructure costs. With the RAPIDS Accelerator for Spark, users can also speed up batch workloads within Dataproc Serverless without provisioning clusters.

Accelerate machine learning inference with NVIDIA AI on Google Cloud Dataflow, a managed service for executing a wide variety of data processing patterns, including both streaming and batch analytics. Users can optimize the inference performance of AI models using NVIDIA TensorRT’s integration with Apache Beam SDK and speed up complex inference scenarios within a data processing pipeline using NVIDIA GPUs supported in Dataflow.

DAccelerate the path to deploy generative AI faster with NVIDIA NIM on Google Cloud Run, a fully managed, serverless compute platform for deploying containers on Google Cloud’s infrastructure. With support for NVIDIA GPUs in Cloud Run, users can leverage NIM to optimize performance and accelerate deployment of gen AI models into production in a serverless environment that abstracts away infrastructure management.

Get easy access to NVIDIA GPU capacity on Google Cloud for short-duration workloads like AI training, fine-tuning, and experimentation using Dynamic Workload Scheduler. With flexible scheduling and atomic provisioning, users can get access to the compute resources they need within services like GKE, Vertex AI, and Batch while enhancing resource utilization and optimizing costs associated with running AI workloads. 

With the NVIDIA Blackwell platform coming to Google Distributed Cloud, enterprises can now securely deploy advanced agentic AI—including Google Gemini models—directly in their own data centers on premises. This integration empowers organizations to harness breakthrough AI performance and scalability for sensitive, regulated workloads while ensuring data privacy, sovereignty, and compliance. By combining the strengths of Google Distributed Cloud and NVIDIA Blackwell, businesses can accelerate innovation with next-generation AI, while maintaining full control over their data and operations.

Google Cloud developed a recipe for disaggregated inferencing with NVIDIA Dynamo, a high-performance, low-latency platform for frontier AI models. This recipe makes it easy to deploy NVIDIA Dynamo on Google Cloud’s AI Hypercomputer, including Google Kubernetes Engine (GKE), vLLM inference engine, and A3 Ultra GPU-accelerated instances powered by NVIDIA H200 GPUs.

The NVIDIA Nemotron family of open models, including Nemotron 3 Nano, will be available on Google Vertex AI Model Garden as NVIDIA NIM microservices. This integration will provide developers and enterprises with access to NVIDIA's leading open-weight models. With a Vertex AI managed deployment, organizations can rapidly develop and deploy custom AI agents powered by Nemotron models while maintaining control over performance, cost, and compliance.

NVIDIA Run:ai Model Streamer comes with native Google Cloud Storage support, supercharging vLLM inference workloads on Google Kubernetes Engine (GKE). This collaboration accelerates loading LLMs from “cold” storage to memory by nearly 5x for faster AI inference—including an example using a 141 GB Meta Llama 3.3-70B model.

Cluster Director is a Google Cloud service designed to optimize the deployment and management of large-scale AI and HPC clusters, accelerated by NVIDIA. This includes full support for large-scale AI systems, including Google Cloud’s A4X and A4X Max VMs powered by NVIDIA GB200 and GB300 NVL72 systems. Google Cloud announced the Preview of Cluster Director support for Slurm on GKE, utilizing SchedMD’s Slinky offering—a company recently acquired by NVIDIA.