Geekbench isn't everything: real workload thread

Synthetic scores mislead for actual server tasks. Here's a concrete test anyone can replicate.

• Provision fresh instance, Ubuntu 24.04
• apt install build-essential
• git clone https://github.com/sqlite/sqlite.git
• Time ./configure && time make -j$(nproc)
• Report: real/user/sys times, CPU model, RAM, provider/region

I've run this across four "free tier" Oracle shapes:

1. VM.Standard.A1.Flex (ARM, 4 OCPU, 24GB): configure 45s, make 3m12s
2. VM.Standard.E2.1.Micro (x86, 1/8th CPU, 1GB): configure 8m30s, make 42m+ (swapped heavily)
3. VM.Standard.E3.Flex (x86, 2 OCPU, 8GB): configure 2m10s, make 8m45s
4. VM.Standard.E4.Flex (x86, 2 OCPU, 16GB): configure 1m55s, make 7m50s

Pro tip: the A1 ARM instance is not just cheaper—it's faster for this workload than x86 shapes costing 4x in burstable credits. Heads up: Oracle's x86 "flex" has variable turbo; run three times and average.

Who else can replicate? Especially curious about non-Oracle ARM (Raspberry Pi 5, Ampere at other providers).

Before anyone runs random git clones and make scripts. What could go wrong:

• The build script could be compromised — verify signatures
• You're exposing compiler toolchain to network — firewall first!
• Oracle's free tier requires inbound rules; default-deny with fail2ban is mandatory
• Time command phones home? No, but `time -v` writes to stderr, check your logging

I ran this in a firejail sandbox on my home server. Results not comparable due to virtualization overhead, but principle stands.

Wesh le server حلو for this test. I tried on my home ARM box, le Raspberry Pi 5 with 8GB, incha'allah it stays up. Configure took 6m20s, make 18m45s. Not bad for 75€ hardware vs. Cloud.

But le VPS-ul from GreenCloudVPS Tokyo, ARM, 2 core, 4GB : configure 3m50s, make 9m10s. Better than Pi despite lower specs, their NVMe is fast. Incha'allah their uptime matches.

So, why are we not talking about overprovisioning? I ran this on HostHatch "dedicated" 4 core, 8GB. First run: make 5m30s. Second run immediately after: 11m20s. Third: 8m45s. Very consistent, no? "Dedicated" my foot. Try three times, report variance! The mean is useless without stddev.