Amazon’s AWS cloud division announced plans to offer customers new chips to build and run artificial intelligence applications, as well as access to Nvidia’s latest chips.
Amazon Web services are trying to position themselves as cloud providers with a variety of cost-effective options. But it’s more than just selling cheap Amazon-branded products. Similar to the online retail marketplace, Amazon’s cloud will be loaded with top-of-the-line products from other vendors, including popular GPUs from top AI chipmakers. Nvidia.
Demand for Nvidia GPUs has skyrocketed since startup OpenAI released its chatbot ChatGPT last year, wowing people with its ability to summarize information and create human-like text. This led to Nvidia’s chip shortage as companies raced to incorporate similar generative AI technology into their products.
Amazon is taking an approach that combines both developing its own chips and giving customers access to Nvidia’s latest chips to take on Microsoft, its biggest competitor in cloud computing. may be helpful. Early this month, microsoft took a similar approach, unveiling its first AI chip, the Maia 100, and saying the Azure cloud will be powered by an Nvidia H200 GPU.
The announcement was made Tuesday at the Reinvent conference in Las Vegas. Specifically, AWS said it provides access to: Nvidia’s The latest H200 AI graphics processing unit. It also announced a new artificial intelligence chip, Trainium2, and a general-purpose Graviton4 processor.
The new Nvidia GPU is an upgrade from the H100, the chip OpenAI used to train its most advanced large-scale language model, GPT-4. Demand for rentals from cloud providers like Amazon is also high as large corporations, startups and government agencies compete for limited supplies of chips. Nvidia says the H200 produces output nearly twice as fast as his H100.
Amazon’s own Trainium2 chip is built to train the kinds of AI models that run AI chatbots like OpenAI’s ChatGPT and its competitors. Startup Databricks and Amazon-backed OpenAI competitor Anthropic plan to build models with a new Trainium2 chip that is four times more powerful than the original model, Amazon said.
The Graviton4 processor is based on the Arm architecture and consumes less energy than other chips. intel or AMD. Graviton4 promises 30% better performance than the existing Graviton3 chip and delivers better output at a price that AWS says. Inflation is higher than normal, prompting central bankers to raise interest rates. Organizations that want to continue using AWS but want lower cloud fees to better manage the economics may want to consider moving to Graviton.
Amazon says more than 50,000 AWS customers are already using Graviton chips.
Finally, as part of its growing relationship with Nvidia, AWS announced that it will operate more than 16,000 Nvidia GH200 Grace Hopper superchips powered by Nvidia GPUs and Nvidia’s Arm-based general-purpose processors. Both Nvidia’s own R&D group and AWS customers will be able to take advantage of this infrastructure.
AWS has released more than 200 cloud products since 2006, when it released EC2 and S3 services for computing and storing data. Not all of them were hits. Some are not updated for long periods of time, and others are retired infrequently, allowing Amazon to reallocate resources. However, the company continues to invest in his Graviton and Trainium programs, suggesting Amazon is sensing the demand.
AWS has not announced a release date for virtual machine instances with Nvidia H200 chips or instances that rely on Trainium2 silicon. Customers can start testing Graviton4 virtual machine instances today, ahead of commercial availability in the coming months.
clock: Deepwater’s Gene Munster says analysts need to raise their growth forecasts for AWS.
