Virtual Machine
VMBack to Glossary
virtualisation
Bare Metal
A physical server provisioned directly to a single workload without a virtualisation layer, delivering maximum compute performance and predictable latency at the cost of reduced density and flexibility.
Full Definition
Bare metal refers to a physical server running without a shared hypervisor layer — the customer's operating system runs directly on the hardware, eliminating virtualisation overhead (typically 5–15% for CPU/memory-bound workloads) and providing consistent, noisy-neighbour-free performance.
Bare metal is preferred for latency-sensitive applications (financial trading, real-time databases), HPC workloads, and AI training where full GPU access without virtualisation overhead is required. Major cloud providers offer bare metal as a service (AWS Bare Metal, Azure BareMetal Solutions, IBM Cloud Bare Metal). Colocation facilities provide standard bare metal through dedicated server hosting.
For AI factory deployments, bare-metal GPU servers are the norm — NVIDIA DGX H100 and GB200 systems ship as bare-metal appliances. Kubernetes runs directly on these bare-metal GPU nodes via the NVIDIA GPU Operator, without requiring a hypervisor layer, maximising throughput for AI training workloads.
Also Known As
dedicated serverbare metal serverphysical server
Source Reference
Uptime Institute Bare Metal vs Virtualisation Performance Analysis; NVIDIA DGX H100 System Guide