How AWS Graviton Achieves Up to 40% Better Price-Performance than x86
AWS Graviton instances can beat the price-performance of comparable x86 cloud servers by over 40%. This allows running applications at lower overall costs or achieving significantly higher throughput per dollar spent.
But how does Graviton provide such an impressive efficiency advantage over generic Xeon or EPYC server CPUs? Let’s examine the architectural innovations that culminate into game-changing improvements in real-world metrics:
Graviton utilizes 64-bit Neoverse cores that incorporate optimizations for maximizing instructions executed per clock cycle. The simplified RISC architecture focuses on efficient data center workloads rather than legacy compatibility.
AWS engineers customized Graviton’s caches, branch prediction and prefetch logic specifically for running scale-out applications. The hardware works hand-in-glove with the OS, libraries and application code for consistent latency.
Leading Edge Manufacturing
Fabricated on the 7nm process, Graviton2 fits over 30 billion transistors on a tiny 250 mm2 die size. Compare this to large 10nm x86 cores still stuck reusing old amortized designs for economies of scale.
Elastic Infrastructure Integration
Instances leverage AWS’s purpose-built infrastructure like nitro cards, elastic networks, EBS storage etc. to achieve unmatched IOPS, throughout and connectivity.
Domain Specific Customization
From number of cores to memory interfaces, Graviton’s architecture aims to excel at targeted use cases like web serving, in-memory caching and container orchestration rather than generic workloads.
Compound Optimization Effects
Each individual optimization builds upon another across full-stack integration of CPU, memory, software and infrastructure. Together they compound into real-world performance per TCO improvements over 40%.
By tailoring the cloud hardware to efficiently handle modern scale-out workloads, AWS Graviton effectively sets a new bar for price-performance – one that generic x86 systems struggle to match. The efficiency lead allows smaller teams to affordably tap into data center-class capabilities on demand.