Cloudflare Gen 13 Servers Double Compute Throughput with AMD EPYC Turin

Cloudflare

Mar 24, 2026 · Updated Apr 25, 2026

Cloudflare deployed Gen 13 servers, doubling compute throughput versus Gen 12 by switching to AMD EPYC Turin 9965 processors. The bet paid off — Cloudflare's FL2 Rust rewrite reduced cache dependency, making Turin a natural fit.

Cloudflare deployed Gen 13 servers built around the AMD EPYC Turin 9965 — a 192-core, 384-thread processor with 2 MB of L3 per core, replacing Gen 12's AMD Genoa-X (96 cores, 1152 MB total L3). Gen 13 also doubles memory to 768 GB DDR5-6400, adds 24 TB PCIe 5.0 NVMe storage, and quadruples network bandwidth to dual 100 GbE.

The cache reduction would have been a problem on FL1 — but Cloudflare's FL2, a Rust rewrite of its request handling layer, does not depend on large shared caches and scales nearly linearly with core count. Running FL2 on Turin 9965 in production, Cloudflare achieved up to 2x throughput versus Gen 12 within latency SLAs, along with 50% better performance per watt.

Teams on Cloudflare Workers and its global edge network will see the benefit as Gen 13 deploys worldwide — twice the compute capacity at current latency budgets.

View the full update on blog.cloudflare.com

Cloudflare

@CloudflareMar 23

Cloudflare’s Gen 13 servers double our compute throughput by rethinking the balance between cache and cores. Moving to high-core-count AMD EPYC ™ Turin CPUs, we traded large L3 cache for raw compute density. By running our new Rust-based FL2 stack, we completely mitigated the latency penalty to unlock twice the performance. https://t.co/1KNDtooLXm

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from Cloudflare →

Keep reading

Cloudflare MCP Server Covers Full API in Two Tools and 1,000 Tokens

Cloudflare launched a new MCP server that covers its 2,500-endpoint API in two tools and ~1,000 tokens. Code Mode lets agents write code to discover and call APIs instead of loading every tool description upfront, cutting token overhead by 99.9%.

Zyphra Launches AMD-First Inference Cloud Optimized for Long-Horizon Agents

ZyphraMay 15

Zyphra Launches AMD-First Inference Cloud Optimized for Long-Horizon Agents

Zyphra launched Zyphra Cloud, a full-stack AI platform on AMD MI355X GPUs rather than NVIDIA, opening with serverless inference for long-horizon agents. The 288GB of memory per AMD chip — versus 192GB on NVIDIA's B200 — keeps nearly double the agent sessions resident in VRAM at long context.

Vercel Uses AI to Rewrite WebStreams for a 14.6x Server-Side Performance Gain

VercelFeb 18

Vercel Uses AI to Rewrite WebStreams for a 14.6x Server-Side Performance Gain

Vercel used AI to autonomously rewrite Node.js WebStream implementation, achieving up to 14.6x performance improvement with 1,100 of 1,116 WPT tests passing. The work is being upstreamed to Node.js, meaning all server-side Node.js workloads benefit once it merges.

Cloudflare adds MiniMax M3 with 1M context for agentic coding

MiniMaxJun 2

Cloudflare adds MiniMax M3 with 1M context for agentic coding

Cloudflare has integrated the MiniMax M3 foundation model into its AI Gateway platform. The update provides developers with a high-context, multimodal model specialized for autonomous coding tasks directly within their existing infrastructure.