Microsoft Upgrades Azure's AI Chops with Nvidia Silicon

Microsoft is leaning heavily on chip giant Nvidia's hardware to advance its AI efforts, as evidenced by the string of announcements made at Nvidia's GTC conference this week.

"Together with NVIDIA, we are making the promise of AI real, helping drive new benefits and productivity gains for people and organizations everywhere," said Microsoft CEO Satya Nadella in a prepared statement on Monday.

The two companies described their joint GTC announcements as expansions of their existing partnership, which in recent years has been focused on creating the right hardware to support AI and machine learning workloads running on Microsoft's infrastructure.

"Through our collaboration with Microsoft," said Nvidia CEO Jensen Huang, "we're building a future that unlocks the promise of AI for customers, helping them deliver innovative solutions to the world."

Notable Microsoft-Nvidia announcements from GTC this week include:

Microsoft Adopting GB200 'Superchips': Nvidia launched its new Blackwell B200 GPU architecture at GTC this week, opening the door to combine it with the older Nvidia Hopper processor to create a "superchip" that can support large-scale generative AI workloads. Microsoft is planting the resulting GB200 processor in its Azure cloud to enable large language model (LLM) training.

"The GB200 will enable Microsoft to help customers scale these resources to new levels of performance and accuracy," noted Azure hardware chief Rani Borkar in a blog post.

Azure is one of the first cloud platforms to offer customers the ability to rent instances powered by the new GB200 chips, which can support LLMs that enable speech recognition, natural language and computer vision.

"Azure customers will be able to use GB200 Superchip to create and deploy state-of-the-art AI solutions that can handle massive amounts of data and complexity," Borkar said, "while accelerating time to market."

New InfiniBand Architecture for Azure AI: Microsoft is also adopting the newly announced Quantum-X800 InfiniBand networking platform to create an "end-to-end AI compute fabric" in Azure. 

Capable of achieving throughput of 800Gb/s, the Quantum-X800 "sets a new standard in delivering the highest performance for AI-dedicated Infrastructure," according to Nvidia's release.

"With new integrations of NVIDIA networking solutions, Microsoft Azure will continue to build the infrastructure that pushes the boundaries of cloud AI," said Microsoft VP Nidhi Chappell in a prepared statement.

Azure AI VMs Powered by Nvidia: Now generally available is the Azure NC H100 v5 series of virtual machines (VMs), which are specifically tuned for generative AI and high-performance computing workloads. They're also well-suited for complex modeling and simulations, particularly for the fields of financial analytics, climate modeling, weather forecasting, quantum chemistry, molecular dynamics and computational fluid dynamics.

Under the hood, the Azure NC H100 v5 VMs are "based on the NVIDIA H100 NVL platform, which offers two classes of VMs, ranging from one to two NVIDIA H100 94GB PCIe Tensor Core GPUs connected by NVLink with 600 GB/s of bandwidth."

Other Microsoft-Nvidia Collaborations Announced at GTC
Omninverse Cloud, Nvidia's digital twins platform, is being released as APIs later this year, with availability first coming to Azure.

The two companies are working to integrate Microsoft Fabric with Nvidia's DGX Cloud, enabling "NVIDIA's workload-specific optimized runtimes, LLMs, and machine learning [to] work seamlessly with Fabric."

Nvidia's generative AI microservices, NIM, will be available on Azure AI.

About the Author

Gladys Rama (@GladysRama3) is the editorial director of Converge360.