Microsoft Corp. wants to entice enterprises to bring their most advanced artificial intelligence projects to the Azure cloud platform, and it’s doing so by offering access to the latest and most powerful graphics processing units from Nvidia Corp.

The company said today its ND H100 v5 Virtual Machine series is generally available through the Azure cloud starting today. It provides customers with access to the powerful computing infrastructure required to train and run generative AI models.

In addition, it’s expanding the availability of its Azure OpenAI Service, through which customers can explore the most advanced AI models from ChatGPT creator OpenAI LP. Azure OpenAI is now available in multiple regions worldwide, the company said.

Microsoft’s ND H100 v5 VMs appears to be one of the most powerful compute offerings it has offered thus far. Available now in Azure’s East United States and West United States regions, they’re equipped with eight of Nvidia’s latest and most formidable H100 GPUs.

Nvidia launched the H100 GPUs last year, explaining that they’re based on its new Hopper architecture. They’re capable of delivering orders of magnitude more processing power than the Nvidia A100 GPUs that were used to train the original ChatGPT.

The advantage of the H100 GPUs is that they deliver “significantly faster AI model performance” than previous-generation GPUs, Nvidia said at the time they were launched. The ND H100 v5 instances also feature Intel Corp.’s latest 4th Gen Intel Xeon Scalable central processing units, with low-latency networking via Nvidia’s Quantum-2 CX7 InfiniBand technology. They also incorporate PCIe Gen5 to provide 64 gigabytes per second bandwidth per GPU, and DDR5 memory that enables faster data transfer speeds to handle the largest AI training datasets.

Microsoft makes some big claims about the performance prowess of the ND H100 v5 instances, promising a six-fold speedup in matrix multiplication operations and 1.8 times faster inference on large language models such as OpenAI’s GPT-BLOOM 175B.

The company said enterprises can register their interest in the new VMs today, and promised it will ramp up the offering to make hundreds of thousands of the H100 GPUs available to customers by next year.

The ND H100 v5 VMs are targeted at generative AI workloads and so it makes perfect sense for Microsoft to also expand the Azure OpenAI Service, which provides direct access to OpenAI’s cutting-edge AI models, GPT-4 and GPT-35-Turbo. Launched in January, the Azure OpenAI service was originally offered only in Azure’s East United States, France Central, South Central United States and West Europe regions, but has now been expanded to Canada East, East United States 2, Japan East and UK South, the company said today.

Holger Mueller of Constellation Research Inc. said today’s announcement shows how enthusiastic Microsoft is about scaling up its AI offerings and making them stronger and more widely available. “It wants to make OpenAI available to more customers and it’s offering Nvidia-based virtual machines alongside it, for customers to run and monetize their custom AI models,” he said. “The speed of the rollout, along with availability and stability, is key for customer adoption, as many enterprises are tied to specific regions due to data residency and security requirements.”

Microsoft said Azure OpenAI Service is already being used by more than 11,000 enterprise customers, expanding at a rate of about 100 new users per day.

“As part of this expansion, we are increasing the availability of GPT-4, our most advanced generative AI model, across the new regions,” said Nidhi Chappell, general manager of Azure HPC, AI, SAP and Confidential Computing at Microsoft. “This enhancement allows more customers to leverage GPT-4’s capabilities for content generation, document intelligence, customer service and beyond.”

Photo: Microsoft