Skip to content
Register

ARM vs x86 for static site hosting: I tested both

VPS Hosting by zoekyo 11 replies 1.5K views
zoekyo

Member OP

Less Is More

#1

I ran nginx on two 2-core VPS at the same price point. One ARM one x86. Same region. Same NVMe. Simple is better, so I expected ARM to win on efficiency.

Benchmarks said ARM should be faster. Geekbench higher. Watt per core lower. But real world? X86 served 12% more requests per second. Static files, no caching, wrk with 4 connections.

I checked the kernel. Same version. Checked nginx config. Identical. The only visible difference: ARM box had a VirtIO driver that looked older. Network card path? Not sure.

Has anyone else seen x86 beat ARM for static hosting despite worse benchmarks

rm -rf / --no-preserve-root ☯
marchhopper

Member

#2
zoekyo said:
The only visible difference: ARM box had a VirtIO driver that looked older

That is a curious result; I have observed similar anomalies with certain VirtIO implementations on ARM machines. The network stack path on those machines often involves more emulation layers than one would expect; this is not always visible in CPU-bound benchmarks. You might check the interrupt distribution across cores on the ARM machine; I have seen IRQ imbalance hide performance on otherwise capable machines. -- Marcus

fanboy

Member

#3

Watt per core is the only metric that matters in 2026. Apple Silicon proved it. Your x86 machine is burning coal for those 12% while ARM sips power. The VirtIO thing is a software problem not an ARM problem. Intel is dying and you are clinging to legacy trash. Fix the driver and ARM crushes x86 at half the watts. This compatibility FUD is old news

Blaze

Member

#4

Have you checked the MTU size! I had a box that crawled because the MTU was wrong and nobody told me! Network cards lie! Benchmarks lie! Only real traffic matters! 12% is 12%! take the win!

kat_fold

Moderator

Moderator

#5

Hey folks, friendly reminder to keep things constructive. @fanboy, calling someone else's setup "trash" isn't helpful. Let's focus on the technical mystery here. 🙂

Zoekyo, did you try iperf3 between the two boxes? That might isolate network from nginx without needing to blame either architecture yet.

~be kind or be gone~
lol3

Member

#6

Yikes that irq thing sounds familiar. I had a box at OVHcloud that lagged hard until I moved irq off cpu0. Gg for testing real traffic not just synthetic. Most noobs just run geekbench and call it a day. Pwn the benchmarks with actual wrk data

TH

Member

#7

I test already my Time4VPS ARM and my Leaseweb x86, same result. The x86 win for small file. I think the network card-driver is bettr on x86,not the cpu. I buy many server-server before, always the driver matter more than benchmark — https://github.com/masonr/yet-another-bench-script. Simple is better, yes

AD

Member

#8

What was the exact nginx version on both?

SS

Member

#9

I had the same IRQ imbalance on Ampere

TA

Member

#10

12% feels small for a real win. Variance?

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft