I noticed odd behavior on my "dedicated" server from Contabo. The lscpu output showed an AMD EPYC 7402, but kernel logs during boot referenced hypervisor hints. dmesg contained "Hypervisor detected: KVM" and /sys/devices/virtual existed. As far as I know, a genuine bare metal machine should not expose these. I opened a ticket. They insisted it was dedicated. I requested the right to audit under GDPR Article 15. They offered a migration credit to a "confirmed bare metal" tier instead. I am documenting this for the community. What reliable detection methods do you use beyond dmesg and lscpu, which can apparently be faked?
My 'dedicated' server was actually a VM with fake lscpu output
1) Detection methods
1.1) CPUID leaf checks
1.1.a) GenuineIntel vs fake strings
1.1.b) Hypervisor bit in ECX
1.2) Timing attacks
1.2.I) rdtsc differences under load
1.2.ii) CPUID execution time
1.3) Hardware inspection
1.3.a) /sys/class/dmi/id/product_name
1.3.b) Presence of virtio devices
2) Emotions about this
2.a) Anger
2.b) Disappointment
2.c) Resignation
3) Actions to take
3.I) Chargeback if misrepresented
3.ii) Report to consumer protection
3.iii) Name and shame on forums
The 23/05/2026 date on your post, Fritz48, matches my own discovery with a different provider. Bloody hell, this is more common than I'd like. The metric: check CPU frequency scaling. A VM often shows identical MHz across all cores regardless of load, because the hypervisor presents a static value. Real silicon varies. Also check for bloody absent MSRs—model-specific registers that KVM does not always pass through. Try rdmsr on 0x1a0, 0x1a2. If you get EINVAL consistently, suspect virtualization. My 2.3 kg server from Vultr turned out to be 0 kg of actual metal. Just electrons pretending.
I had a "dedicated" box from OVHcloud in 2021. Ran for eight months. One morning I lost all connectivity. Provider said "hardware failure, migrating to new node." Took six hours. Came back with different MAC addresses, different CPU stepping in /proc/cpuinfo, and my ZFS pool import failed because the "new" machine had a different virtual disk geometry. They were running my "dedicated" server as a VM snapshot. Verify your restores weekly. Also verify your hardware monthly. I now keep a bootable USB with lspci -vv output from day one. Paranoid? I have the USB.
Hi Fritz48, this happened to me too with Hostinger! The server they sold me, VPS price but "dedicated" label, I run the kvm-ok command, it says acceleration can be used—why on bare metal? I check the dmesg, see the virtio everywhere, They offer me 3 months credit, I say no, I move to InterServer — https://www.interserver.net — real dedicated there, You must fight, The provider must not win,
I don't trust that "confirmed bare metal" tier either
RAJ you may be right. I asked Contabo specifically: which datacenter, which rack, may I send my own technician. They said "for security reasons we do not disclose physical location." GDPR Article 15 gives right to know processing location. They stalled. I am now talking to a lawyer in Berlin about filing with Bundesnetzagentur.
I am on Vultr bare metal in Stockholm. Real server, took 40 minutes to provision which felt right for someone walking to rack. Lspci shows actual Intel NICs, not virtio-net. But I pay 120 USD. You cannot get bare metal for VPS prices. The market has spoken.
This. My solar setup cost more than my first car. Real hardware has real costs. The 4 EUR Contabo tier is not magic. It is arithmetic. Someone pays the difference, or someone lies.
My 1% utilization VPS at Hostinger is honest about what it is. I pay 6 USD, I know I share. The problem is not VM. The problem is lie.