BotBeat
...
← Back

> ▌

CloudflareCloudflare
UPDATECloudflare2026-03-23

Cloudflare Achieves 2x Throughput and 50% Better Power Efficiency with AMD EPYC Turin Upgrade

Key Takeaways

  • ▸Cloudflare's Gen 13 servers based on AMD EPYC 9965 processors deliver 2x the throughput of previous Gen 12 infrastructure
  • ▸Software optimization through FL2's Rust-based rewrite of request handling achieved up to 50% more requests per CPU and 70% lower latency
  • ▸The new platform offers 50% better performance-per-Watt efficiency and up to 60% higher rack throughput, improving both performance and sustainability
Source:
Hacker Newshttps://www.phoronix.com/news/Cloudflare-Gen13-Server-Turin↗

Summary

Cloudflare has unveiled its new Gen 13 server platform built around AMD EPYC 9005 "Turin" processors, delivering significant performance and efficiency improvements over its previous generation infrastructure. The upgrade achieves 2x throughput compared to Gen 12 Genoa-X servers, 50% better performance-per-Watt, and up to 60% higher rack throughput. The gains come from both hardware advancements in the EPYC 9965 flagship processor and software improvements in Cloudflare's FL2 framework, a Rust-based rewrite of their core request handling layer. FL2's modern architecture optimizations enabled up to 50% more requests per CPU and up to 70% lower latency, while AMD's Platform Quality of Service (PQOS) extensions provided better control over shared resources like cache and memory bandwidth.

  • AMD PQOS extensions enabled fine-grained control over shared resources, contributing to the efficiency gains
MLOps & InfrastructureAI Hardware

More from Cloudflare

CloudflareCloudflare
RESEARCH

Cloudflare Rethinking Cache Architecture for AI-Driven Traffic Era

2026-04-02
CloudflareCloudflare
PRODUCT LAUNCH

Cloudflare's Workers AI Enters Large Model Inference Market With Moonshot AI's Kimi K2.5

2026-04-02
CloudflareCloudflare
PRODUCT LAUNCH

Cloudflare Slashes AI Agent Token Costs by 98% With RFC 9457-Compliant Error Responses

2026-03-27

Comments

Suggested

Google / AlphabetGoogle / Alphabet
RESEARCH

Deep Dive: Optimizing Sharded Matrix Multiplication on TPU with Pallas

2026-04-05
NVIDIANVIDIA
RESEARCH

Nvidia Pivots to Optical Interconnects as Copper Hits Physical Limits, Plans 1,000+ GPU Systems by 2028

2026-04-05
Independent ResearchIndependent Research
RESEARCH

New Research Proposes Infrastructure-Level Safety Framework for Advanced AI Systems

2026-04-05
← Back to news
© 2026 BotBeat
AboutPrivacy PolicyTerms of ServiceContact Us