AWS Graviton5's improved design increases speed and energy efficiency (www.amazon.science)

Amazon Web Services (AWS) has unveiled its latest CPU, the Graviton5, which boasts significant advancements in performance and energy efficiency. This new chip features a sophisticated four-chiplet architecture and supports DDR5-8800 memory along with PCIe gen6 interconnects, enhancing performance by 25% for general-purpose and agentic AI workloads. With 192 cores—double that of the Graviton4—and 192 MB of L3 cache, Graviton5 offers improved branch prediction via its Neoverse V3 core, resulting in up to 30% better execution for real applications like databases. The implications of the Graviton5's design are substantial for the AI and machine learning community. Its three-nanometer process improves circuit density and on-chip communication, while a custom die-to-die connectivity delivers up to 420 GB/s bandwidth between chiplets, decreasing memory latency. Moreover, the introduction of the Nitro Isolation Engine enhances security by providing a formally verified environment that ensures isolation between virtual machines. This combination of improved computational capabilities, energy efficiency, and security positions AWS Graviton processors as a formidable choice for various workloads, advancing the state of cloud computing and AI infrastructure.