Cloudflare's Gen 13 servers: trading cache for cores for 2x performance

Cloudflare goes core‑crazy for 2x speed — readers demand receipts and real benchmarks

TLDR: Cloudflare claims double the speed with new AMD-powered servers and a Rust rewrite, trading “smart memory” for more cores. Commenters aren’t buying it without split benchmarks, roasting the blog’s marketing tone and sparking a Rust vs LuaJIT debate — making transparency the real headline.

Cloudflare just rolled out its Gen 13 servers, bragging about 2x edge performance by ditching big on‑chip “memory” (cache) for a ton more CPU muscle — think fewer brainy shortcuts, way more brawn. The secret sauce, they say, is a rewrite of their request handler into FL2, built in Rust, which supposedly lets performance scale with all those new AMD EPYC Turin cores at the network’s “edge” (servers close to you).

But the comments? Spicy. One reader flat-out calls the post light on details, saying the fix sounds like “we rewrote it in Rust” — cool story, where’s the data? Another blasts the blog’s marketing tone as “slop,” complaining the structure raises more questions than answers. The loudest chorus: show us split benchmarks — how much is the new hardware versus the new software? As one commenter puts it, give us FL2 on the old Gen 12 vs FL1 on the same box. Without that, the 2x claim feels like vibes.

Meanwhile, a cheeky nostalgia thread ignites: someone compares this to the old Celeron vs Pentium days — fewer smarts, higher clocks, surprise wins, plus overclocking lore. And a stealth hot take lands: maybe the “unspoken takeaway” is just how fast LuaJIT (the old stack’s likely tech) already was, even next to Rust. Translation: the community is torn between applauding the upgrade and side‑eyeing the storytelling. They want receipts, charts, and fewer colons in section headers. Until then, it’s Rust vs LuaJIT vs Marketing Blog — fight night at the edge.

Key Points

  • Cloudflare launched Gen 13 edge servers based on AMD EPYC 5th Gen Turin processors, aiming for up to 2x performance.
  • A Rust-based rewrite of the request handling layer (FL2) reduced cache dependence, enabling scaling with higher core counts while maintaining SLAs.
  • Turin offers up to 192 cores (384 threads with SMT), improved IPC via Zen 5, better per-core power efficiency (up to 32% fewer watts than Genoa‑X), and DDR5‑6400 support.
  • High-density Turin parts prioritize throughput over per-core cache: 192 cores share 384 MB L3 (~2 MB/core), versus Gen 12 Genoa‑X’s 12 MB per core with 3D V‑Cache.
  • Gen 12 used AMD Genoa‑X 9684X (96C/192T, 12 MB L3/core); an example Gen 13 option is AMD Turin 9755 (128C/256T, 4 MB L3/core).

Hottest takes

"I don’t think they explained how they solved the cache issue except to say they rewrote the software in Rust, which is pretty vague." — gdwatson
"Nah I'm good thanks. Slop takes more effort to read and just raises questions of accuracy." — danpalmer
"the unspoken takeaway is just how shockingly performant LuaJIT is, even relative to Rust." — alberth
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.