Gen 13 Hardware Foundation
Cloudflare has launched its 13th generation server infrastructure built on AMD EPYC 5th Generation Turin-based processors, delivering significant edge compute improvements. The new generation features up to 192 cores (384 threads) compared to Gen 12's 96 cores, representing a 2x increase in raw core count. Beyond core scaling, Turin processors deliver Zen 5 architectural improvements for better instructions-per-cycle performance, 32% better power efficiency per core, and DDR5-6400 memory support to feed the expanded processor cores.
The Cache-Throughput Tradeoff Challenge
The transition required solving a fundamental architectural challenge. Turin processors deliberately trade per-core L3 cache for higher density and throughput—each core has access to just 2MB of L3 cache, compared to Gen 12's 12MB per core allocation. This 6x cache reduction created a critical problem: when Cloudflare's legacy FL1 request handling layer (NGINX/LuaJIT-based) ran on Turin hardware, L3 cache miss rates spiked dramatically, forcing expensive DRAM memory fetches (350+ cycles versus 50 cycles for L3 hits) and causing latency penalties exceeding 50% at high utilization—unacceptable for production services.
FL2 Software Stack Enables Hardware Potential
The solution came through Cloudflare's FL2 transition, a complete Rust-based rewrite of the core request handling layer that eliminates reliance on large per-core caches. By optimizing cache-agnostic code patterns, FL2 allows performance to scale linearly with core count while maintaining strict latency SLAs. This architectural redesign proved critical: Gen 13 now delivers 2x edge compute performance without the latency penalties that plague cache-dependent workloads on higher-density processors.
Performance Optimization Strategy
Cloudflare collaborated with AMD to tune Gen 13 deployments, leveraging AMD Platform Quality of Service (PQOS) extensions to allocate dedicated Core Complex Dies (CCDs) exclusively to FL2 workloads. This approach maintains cache coherency within isolated processor regions while preventing resource contention. Additional optimizations include NUMA-aware core affinity and strategic workload isolation configurations that balance throughput gains with latency requirements.
Developer and Operational Impact
The Gen 13 deployment directly benefits Cloudflare Workers and edge applications that run on the platform. Developers deploying serverless functions to Cloudflare's network will experience improved request processing throughput and more consistent latency characteristics. From an infrastructure perspective, the higher core density and improved power efficiency per core improve total cost of ownership, enabling broader edge compute capacity across Cloudflare's global network.