Amazon Launches Trainium2 & Graviton4 to Rival Microsoft

On November 28th, Amazon made a significant announcement, unveiling the Trainium2 AI chip and Graviton4 processor for its cloud computing service, AWS. This strategic move aims to address the escalating competition in the AI market, particularly against Microsoft's dominance.

Adam Selipsky, the CEO of Amazon AWS, introduced Trainium2, the second-generation chip designed explicitly for training AI systems, during an event in Las Vegas. Boasting a fourfold increase in performance and double the energy efficiency compared to its predecessor, this chip can be seamlessly deployed within the EC2 UltraCluster, supporting clusters of up to 100,000 chips. This enables the efficient training of foundational models and large-scale language models.

In addition to Trainium2, Amazon also introduced its fourth custom processor, Graviton4. With a 30% performance improvement, 50% more cores, and a 75% increase in memory bandwidth compared to its predecessor, Graviton4 ensures optimal cost-effectiveness and energy efficiency for a wide range of workloads on Amazon EC2 cloud services. Notably, the processor enhances security by encrypting all high-speed physical hardware interfaces.

This move by Amazon follows recent developments from major tech players such as Microsoft with its Maia AI chip and Google's Tensor Processing Unit (TPU). The introduction of custom AI chips by these industry leaders reflects a strategic shift away from heavy reliance on NVIDIA, meeting the evolving demands of technologies like large language models.

Selipsky disclosed that AWS is set to offer the new training chips starting in 2024. It's important to note that while AWS will embrace these innovations, the cloud services will continue to provide offerings based on the latest chips from NVIDIA.

Significantly, both Amazon and Microsoft have adopted the Arm architecture for their custom chips, aligning with a broader industry trend of diversifying away from traditional reliance on processors manufactured by Intel and AMD. Notably, Oracle has also incorporated chips from the startup Ampere Computing into its cloud services.

This move by Amazon is poised to make a lasting impact on the landscape of AI chip technology, underscoring the company's commitment to advancing the capabilities of its cloud computing services and staying at the forefront of technological innovation.