NVIDIA Blackwell GB200 Enhances Google Cloud A4 VMs with 2.25x Compute Power and Increased HBM Capacity

Google Cloud has achieved a significant advancement in AI training and development with the introduction of its first NVIDIA Blackwell “GB200 NVL72″system, specifically designed to enhance OpenAI’s performance.

Expansion of NVIDIA Blackwell GB200 Systems in Google Cloud and Azure AI Service

OpenAI is set to experience a substantial increase in computational capabilities due to the deployment of NVIDIA’s GB200 NVL72 “Blackwell”system. While NVIDIA has long supported OpenAI with its advanced AI models, this latest integration with Google Cloud signifies a crucial leap forward in leveraging powerful data center GPUs for improved AI training.

NVIDIA Blackwell is here!

Today, we’re excited to bring the highly-anticipated @nvidia Blackwell GPUs to Google Cloud with the preview of A4 VMs, powered by NVIDIA HGX B200.

Learn more → https://t.co/N1oqZMzJ8M pic.twitter.com/iBTLqUQSVg

— Google Cloud Tech (@GoogleCloudTech) January 31, 2025

In addition to Google Cloud, Microsoft has also unveiled its initial Azure instances featuring NVIDIA’s cutting-edge Blackwell GPUs. In a recent X post, OpenAI’s CEO, Sam Altman, highlighted the deployment of the complete eight racks of the GB200 NVL72 system within Microsoft Azure. Altman expressed gratitude to both Microsoft CEO Satya Nadella and NVIDIA CEO Jensen Huang for facilitating this collaboration, which positions Microsoft as a leading cloud service provider for AI-driven workloads.

first full 8-rack GB200 NVL72 now running in azure for openai—thank you @satyanadella and jensen!

— Sam Altman (@sama) January 31, 2025

The state-of-the-art NVIDIA Blackwell GB200 NVL72 configuration consists of 36 Grace CPUs and 72 Blackwell GPUs arranged in a rack-scale architecture. With a total of eight racks implemented, the system provides an impressive capability of 288 Grace CPUs and an astounding 576 B200 GPUs. This setup can deliver remarkable computational prowess, with each rack capable of generating up to 6, 480 TFLOPS of FP32 and 3, 240 TFLOPS of FP64 performance. When aggregated across eight racks, the total potential rises to approximately 51, 840 TFLOPS of FP32 and 25, 920 TFLOPS of FP64.

This exceptional computational capability makes the Azure OpenAI service an invaluable asset for enterprises seeking to enhance their operational efficiency. The NVIDIA “Blackwell”GB200 NVL72 is specifically engineered for large-scale generative AI workloads, boasting an impressive memory bandwidth of up to 576 TB/s while facilitating advanced parallel processing.

Currently, OpenAI utilizes a variety of NVIDIA GPU models, including V100s, H100s, and A100s; however, the Blackwell B200 emerges as the most sophisticated GPU chip, offering substantial performance enhancements over its predecessors. With Microsoft’s investment of nearly $14 billion in OpenAI, it’s evident that the company aims to provide its customers with access to the most powerful machines available on the market.

Source & Images