Microsoft teams up with Nvidia for a new supercomputer
Microsoft is working on a new supercomputer with Nvidia. The multi-year deal comes two years after Microsoft partnered with OpenAI to build the same type of technology
Microsoft and Nvidia aim to create “one of the most powerful AI supercomputers in the world”.
The US chip design and computer company has announced a multi-year partnership with the Redmond-based tech giant to build a "huge" computer to handle AI computationally intensive work in the cloud.
The AI computer will run on Microsoft's Azure cloud, using tens of thousands of graphics processing units (GPUs), Nvidia's more powerful H100 and its A100 chips. Nvidia declined to say how much the deal is worth. But industry sources estimate the A100 chip is priced at around $10,000-12,000, and the H100 graphics processing unit (GPU) is even more expensive, Reuters reveals.
About two years ago, Microsoft announced a partnership with OpenAI, the artificial intelligence lab with which it has a close business relationship, to build what the tech giant called an "AI supercomputer" running in the Azure cloud. With over 285,000 processor cores and 10,000 graphics cards, Microsoft claimed at the time to be one of the largest supercomputer clusters in the world.
But now the partnership with Nvidia presumably comes to support even more ambitious AI workloads, Techcrunch comments.
The rapid growth of AI models such as those used for natural language processing has greatly increased the demand for faster and more powerful computing infrastructures. “A recent study found that the computational requirements for large-scale AI models doubled at an average rate of 10.7 months between 2016 and 2022,” adds the American newspaper.
All the details.
NVIDIA'S STRATEGY
"We're at that inflection point where AI is coming to enterprise and it's becoming real to get those services that customers can use to implement AI for business use cases," Ian Buck, director of AI, told Reuters . Nvidia general for Hyperscale and HPC. “We are seeing widespread adoption of AI and the need to apply AI to enterprise use cases.”
In addition to selling the chips to Microsoft, Nvidia said it will work with the software and cloud giant to develop AI models.
Nvidia said Azure would be the first public cloud to use its Quantum-2 InfiniBand networking technology which has speeds of 400 gigabits per second. That networking technology connects servers at high speeds. This is important as the heavy computation work of the AI requires thousands of chips to work together across multiple servers.
THE POWER OF THE H100 GPU
In particular, the H100, the flagship of Nvidia's Hopper architecture, comes with a special "Transformer Engine" to accelerate machine learning tasks and, at least according to Nvidia, offers 1.5 to 6x performance better than the A100. It's also less power hungry, delivering the same performance as the A100 with up to 3.5x better energy efficiency.
“One of the first utility-scale machines to sport H100 GPUs, the Henri system built by Lenovo and operated by the Flatiron Institute in New York City, topped this year's list of most efficient supercomputers,” reports TechCrunch .
MICROSOFT'S GOAL
"Our partnership with Nvidia unlocks the world's most scalable supercomputing platform, delivering cutting-edge AI capabilities for every enterprise on Microsoft Azure," said Scott Guthrie, executive vice president of Microsoft's cloud and AI group, in a statement.
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/microsoft-si-allea-con-nvidia-per-un-nuovo-supercomputer/ on Fri, 18 Nov 2022 06:45:11 +0000.