- Amazon Web Services CEO Adam Selipsky revealed Trainium2, a cutting-edge chip tailored for training AI models, during the Collision conference in Toronto on June 27, 2023. Alongside this announcement, AWS will grant access to Nvidia’s upcoming H200 Tensor Core GPUs.
- Strengthening their collaboration, AWS has allocated a specialized computing cluster for Nvidia’s utilization.
- Currently, AWS clients can explore and evaluate the new versatile Graviton4 chips, which offer improved performance and efficiency.
At the Reinvent conference in Las Vegas, Amazon’s AWS cloud division introduced the Trainium2 AI chip and the versatile Graviton4 processor. In conjunction with these launches, AWS unveiled its partnership with Nvidia to provide access to the state-of-the-art H200 AI GPUs.
In a bid to establish itself as a premier cloud service provider, Amazon Web Services is concentrating on delivering a diverse array of cost-effective solutions. This strategy revolves around offering top-notch products, including coveted GPUs from Nvidia, aligning with the company’s stellar reputation in online retail and cloud services.
By introducing the Graviton4 processors, which leverage Arm architecture and tout superior energy efficiency compared to Intel or AMD chips, AWS aims to deliver a 30% performance boost over the existing Graviton3 chips. This advancement not only enhances productivity but also tackles prevailing economic challenges by furnishing a more budget-friendly cloud solution for enterprises.
With over 50,000 AWS customers already leveraging Graviton chips, the debut of the Trainium2 chips pledges a fourfold enhancement in performance compared to their precursor. Startups like Databricks and Anthropic, supported by Amazon and a rival of OpenAI, are poised to harness these chips for constructing advanced models.
AWS intends to deploy more than 16,000 Nvidia GH200 Grace Hopper Superchips, featuring H100 GPUs and Nvidia’s Arm-based processors, exclusively for Nvidia’s research and development initiatives. Meanwhile, the surging demand for Nvidia GPUs, catalyzed by the triumph of OpenAI’s ChatGPT chatbot, has resulted in a scarcity of these chips as companies hasten to integrate generative AI technologies into their offerings.
Despite the potential competitive threat posed by AWS’s AI chip offerings to Nvidia, Amazon is expanding its partnership with the chipmaker. This strategic maneuver furnishes AWS customers with an additional AI computing alternative, particularly beneficial for those unable to acquire the latest Nvidia GPUs.
As a trailblazer in cloud computing, Amazon has been providing GPU rentals for over a decade. The introduction of in-house-developed AI processors in 2018, following suit with other cloud rivals, aimed to deliver robust computing capabilities at a reasonable cost, further cementing Amazon’s foothold in the cloud landscape.
While AWS has rolled out numerous cloud products since its establishment in 2006, the emphasis remains on advancing the Graviton and Trainium initiatives, signaling robust market demand for these innovative solutions. Customers can already commence testing Graviton4 instances, with commercial availability slated in the upcoming months, while rollout dates for virtual-machine instances featuring Nvidia H200 chips and Trainium2 silicon are yet to be disclosed.
