Cloudflare just rolled out their newest generation of servers—Gen 13—and the headline number is impressive: 2x the performance of the previous generation. But the fascinating story isn’t the hardware upgrade. It’s what happened when that hardware initially made everything worse, and how they fixed it.
This story matters for every business leader thinking about infrastructure upgrades, because it illustrates a crucial truth: sometimes the path to better performance requires tearing down and rebuilding your foundation.
The Hardware Looks Perfect on Paper
Cloudflare’s Gen 13 servers use AMD’s latest EPYC Turin processors—monsters with 192 cores (double the previous generation). More cores mean more work happening simultaneously. More memory bandwidth. More storage. On paper, it’s a slam dunk.
But when they fired up these new servers with their existing software stack, something alarming happened: latency spiked by over 50%. The new “faster” servers were making websites load noticeably slower.
The Cache Trap
Here’s what went wrong, in non-technical terms:
The previous generation servers (Gen 12) had massive onboard cache memory—think of it as a super-fast notepad sitting right next to the processor. When your website needed information, the processor could grab it instantly from this notepad without traveling across the room to the filing cabinet (main memory).
The new processors (Gen 13) traded that massive notepad for more hands to do work. Instead of 96 workers with huge notepads, you now had 192 workers with tiny notepads. Cloudflare’s software was designed around those big notepads—and suddenly, everyone was constantly running back to the filing cabinet.
More processing power was being wasted on walking back and forth.
The Gutsy Solution: Rewrite Everything
Cloudflare faced a choice many businesses encounter with infrastructure upgrades:
- Retreat: Go back to the old servers and wait for processors that match your software’s assumptions
- Work around it: Try to optimize the existing software to work better with the new limitations
- Rebuild: Fundamentally redesign your software to match the new hardware’s strengths
They chose option 3—the hardest path. They rewrote their core request-handling layer (called FL1, now FL2) from the ground up in a different programming language (Rust), specifically designed to work efficiently with less cache but more cores.
The result? Not just the 2x performance they expected—but 70% lower latency and 50% better power efficiency. They turned a potential disaster into their best server generation yet.
Why This Matters for Your Business
This isn’t just a cool tech story. It’s a lesson about infrastructure investment that applies to businesses of all sizes:
1. Hardware upgrades don’t automatically mean better performance. Your software, processes, and architecture need to match your infrastructure. Buying faster servers while running outdated software is like buying a sports car and only driving it in first gear.
2. Sometimes the right move is to rebuild, not patch. When Cloudflare’s existing software hit its limits, they didn’t just optimize around the edges. They acknowledged the old approach had reached its end and rebuilt with a modern foundation. Many businesses cling to outdated systems too long, trying to squeeze out “just a little more” performance.
3. Short-term pain can prevent long-term stagnation. The rewrite was expensive and time-consuming. But now Cloudflare has a foundation that will serve them for years—and competitors still patching old architecture will struggle to catch up.
4. Efficiency = cost savings at scale. Cloudflare’s 50% better performance-per-watt doesn’t just help the planet—it directly reduces operational costs. For any business running servers (whether on-premises or in the cloud), efficiency improvements compound into significant savings.
The Practical Takeaway
You don’t need to be operating at Cloudflare’s scale for these lessons to apply. Every business with aging infrastructure faces similar decisions:
- Is your current system holding you back from modern capabilities?
- Are you spending more time working around limitations than delivering value?
- Would rebuilding on a modern foundation pay off over the next 3-5 years?
The companies that thrive are those willing to make strategic rebuilds when the fundamentals have shifted—not just for the sake of being modern, but because the old approach genuinely can’t deliver what the business needs.
Want to evaluate whether your infrastructure is ready for your business’s next phase—or holding you back? Let’s talk. At Uptown4, we help businesses make smart infrastructure decisions that balance immediate needs with long-term scalability.