-
Microsoft announces its first artificial intelligence accelerator chip
-
Redmond is working with AMD and Nvidia to accelerate AI for Azure
-
Microsoft and Nvidia are using Azure to customers to develop their own large language models
Artificial intelligence (AI) is the future.
We’re on a tipping point in this new age of AI.
These are familiar phrases you have heard before, and are probably sick of, by now.
Nonetheless, Microsoft CEO Satya Nadella laid them on with abandon during the opening keynote to the firm’s Ignite event this week.
He brought some heavy hitters, such as Nvidia CEO Jensen Huang, onstage to talk about the AI and cloud work Nvidia has already done with Microsoft.
Nadella said that Nvidia is making an “AI foundry” available on Microsoft’s Azure public cloud “called DGX cloud.”
Huang said Nvidia, with Microsoft, wanted to “help customers build their own large language models (LLMs).”
“We are going to do for people who want to build their own proprietary [LLMs] what TSMC does for us, we’ll be a foundry for AI models,” he said.
So far, Silverlinings has seen that a only few large organizations, notably telco titans Deutsche Telekom and SKT, that actually want to build their own LLMs. So far, most enterprises want to use ready-made LLM models for AI.
But as the Microsoft CEO said: “We’re just getting started.”
Chips and bits
Nadella said that Microsoft is using AI accelerators from Nvidia and AMD with Azure as the AI workloads grow. “We’re adding Nvidia’s latest GPU AI accelerator H200...to support even larger model inferencing with the same latency.”
The Redmond ruler also welcomed AMD into the AI cloud cuddle. “AMD’s flagship M300X AI accelerator is coming to Azure to give us even more choice for AI-opmitized VMs,” Nadella said. “This means we can serve large models using fewer GPUs.”
As well as revealing Microsoft’s new Cobalt CPU at the event, Nadella announced that Microsoft is itself getting into the AI accelerator game. Nadella said that Azure Maia will be Microsoft’s “first, fully custom AI accelerator.”
“Starting with Maia 100 designed for running cloud AI workloads like LLM training and inference,” Nadella said.
The CEO recognized that the cooling demands of the new chip will also require new techniques, and have designed a “Sidekick” cooling unit for the chip that can be installed in a rack in a data center.
Nadella said that Microsoft will roll out Maia to support it’s own workloads first, although he didn’t say exactly when this will start.
Want to discuss AI workloads, automation and data center physical infrastructure challenges with us? Meet us in Sonoma, Calif., from Dec. 6-7 for our Cloud Executive Summit. You won't be sorry.