Amazon Web Services (AWS) is making significant strides in the artificial intelligence (AI) landscape, announcing the Trainium2 chip for AI model training and revealing plans to provide access to Nvidia’s cutting-edge H200 Tensor Core graphics processing units. The move positions AWS as a formidable player in the cloud computing sector, offering a diverse range of cost-effective options beyond its proprietary products.
In a bid to outshine competitors and address the surging demand for Nvidia graphics processing units (GPUs) sparked by OpenAI's ChatGPT, AWS adopts a dual strategy. The company is not only developing its own AI chips but also facilitating customer access to Nvidia's latest chips. This strategy is a strategic response to Microsoft's similar approach, which recently introduced its AI chip, the Maia 100, and announced compatibility with Nvidia H200 GPUs in its Azure cloud.
The announcements were made at the Reinvent conference in Las Vegas, where AWS outlined plans to provide access to Nvidia's H200 AI GPUs, unveiled the Trainium2 AI chip, and introduced the general-purpose Graviton4 processor.
The latest Nvidia GPU, an upgrade from the H100, offers enhanced capabilities, promising output almost twice as fast. Given the high demand for these GPUs, AWS aims to meet customer needs by making them available for rent through its cloud services.
Amazon's Trainium2 chips are specifically designed for training AI models, catering to applications like OpenAI's ChatGPT. With an emphasis on improved performance, these chips are expected to be utilized by prominent entities such as startup Databricks and Amazon-backed Anthropic.
The Graviton4 processors, built on Arm architecture, boast energy efficiency compared to Intel or AMD chips. With a promise of 30% better performance than their predecessors, Graviton4 chips provide a cost-effective solution for AWS customers seeking optimal output within budget constraints. Over 50,000 AWS customers are already leveraging Graviton chips, highlighting their popularity.
As part of its expanded collaboration with Nvidia, AWS will operate over 16,000 Nvidia GH200 Grace Hopper Superchips. This infrastructure, incorporating Nvidia GPUs and Arm-based processors, will be accessible to Nvidia's research and development group and AWS customers alike, strengthening the partnership between the two tech giants.
Despite launching over 200 cloud products since 2006, AWS continues to invest in the Graviton and Trainium programs, indicating a sustained commitment to meeting evolving market demands. While specific release dates for virtual-machine instances featuring Nvidia H200 chips and Trainium2 silicon were not disclosed, AWS customers can begin testing Graviton4 virtual-machine instances, expected to be commercially available in the coming months. This strategic move positions AWS at the forefront of AI innovation, catering to the growing needs of diverse industries relying on advanced computing capabilities.
