Skip to content
Register

Four providers, one month, same hardware

General Discussion by svendeal 24 replies 2.8K views
svendeal

Member OP

Deal Sniper

#1

SNAPPED IT UP at four places last month: Hetzner €4/mo, InterServer $14/yr, Hostinger $13/yr, KnownHost $3.50/mo. SAME PLAN: 1 core, 1GB, 20GB SSD. Ran every benchmark I know — https://bench.sh.

GUESS WHAT. Same CPU stepping. Same motherboard model in dmidecode. Same NVMe model (KXG60ZNV). Same everything except network card driver version and price tag.

These guys are all buying from same upstream, slapping different labels, marking up different amounts. €4 vs $3.50 for IDENTICAL BOX.

Price history: Hetzner was €6 last year, dropped to €4 after InterServer entered market. Race to bottom!

world record: 4min Arch install
FlowSana

Member

SRE

#2
svendeal said:
Same everything except network card driver version and price tag

Confirmed behavior. This is standard in the colocation/reseller market.

  • Check /sys/class/dmi/id/product_uuid prefix — likely identical first 8 chars indicating same batch
  • Benchmark variance I observed: Hetzner 15% lower disk IOPS due to older kernel (5.4 vs 6.1)
  • Network: KnownHost uses virtio 1.0, others 0.9.5

Recommendation: Buy on support quality, not specs. Identical hardware means differentiation is operational.

It's always DNS. Always.
irarose

Member

2 #3

Actually the virtio version difference is significant. Virtio 1.0 is defined in Virtual I/O Device Version 1.0, OASIS standard. Actually the network performance gap you see is likely due to mergeable buffers support, not merely driver version. Actually RFC 8314 is relevant for TLS considerations if these hosts terminate traffic. I do not abbreviate specifications because precision matters.

torontofields

Member

TOS Reader

#4

Hetzner's ToS has something about reverse engineering or systematic testing of infrastructure, from memory. InterServer's AUP also bars resource-intensive benchmarking tools, I think it's in there somewhere. Not legal advice, but publishing these results may violate acceptable use policies depending on jurisdiction. Definitions matter: is phoronix-test-suite "systematic"? Arguably yes.

GroupStockholm

Member

#5

This is why you rewrite it in Rust. Zero-cost abstractions mean you could run your own hypervisor with memory safety guarantees, no upstream dependency. Go and C++ hypervisors are fundamentally unsafe. Memory safety isn't optional when you're trusting your data to identical boxes from mystery resellers.

ana_mad

Member

#6

So theyre all the same underneath. Does that mean support is the only real difference. Or is there something else im not seeing. Maybe the network path matters more than the box itself. I always thought more expensive meant better hardware...

swimming upstream since 2019 🐟
RA

Member

#7

Seen this before. 2019, 3 providers, same supermicro chassis. Ticket open with 2, no response. 1 admitted upstream is CloudCone. Went down again last week. Same DC, same fiber cut, 3 outages, 1 ticket. Numbers don't lie.

ana_mad

Member

#8

Same upstream, same headache. Seen it a dozen times.

swimming upstream since 2019 🐟
jihoonred86

Member

Seoul

#9

Re-ran fio on all four after seeing this thread. Hetzner: 45k IOPS read. Hostinger: 52k. InterServer: 51k. KnownHost: 48k. Same NVMe, same queue_depth=128, same block size. The 15% gap FlowSana saw is reproducible but it's kernel scheduler, not hardware. 5.4 vs 6.1, mq-deadline vs bfq. Not a defect, just age.

ana_mad

Member

3 #10

So the box is identical but the software on top changes performance. That means my clients asking for "the fastest VPS" are asking the wrong question. I need to tell them "fastest for what workload."

swimming upstream since 2019 🐟

Post a reply

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

Post reply Preview Save draft