Bare-Metal Compute

The metal, without the middleman.

Direct hardware access with zero hypervisor tax. Our bare-metal instances deliver 100% of the CPU, RAM, and NVMe throughput to your workload — no noisy neighbors, no performance caps, no surprises.

Starting from $6/mo · Provisioned in < 60 seconds · NVMe included

Why bare-metal

Infrastructure that gets out of your way.

0%

Virtualization overhead

No hypervisor layer, no KVM overhead, no memory balloon driver stealing cycles from your workload.

100%

NVMe IOPS delivered

PCIe 5.0 NVMe with direct attachment — every read and write hits the hardware without a virtual storage driver in the path.

<60s

Provisioning time

Pre-imaged hardware and a zero-touch bootstrap pipeline get your instance live in under a minute from first API call.

Instance types

Pick your hardware profile.

All instances run on AMD EPYC™ 9004-series processors with PCIe 5.0 NVMe storage. Bandwidth is measured as monthly egress included in the base price.

NAMEvCPURAMSTORAGEBANDWIDTHPRICE/MO
nb-metal-1
4 vCPU8 GB80 GB NVMe2 TB
nb-metal-2
8 vCPU16 GB160 GB NVMe4 TB
nb-metal-4Popular
16 vCPU32 GB320 GB NVMe6 TB
nb-metal-8
32 vCPU64 GB640 GB NVMe8 TB
nb-metal-16
64 vCPU128 GB1.2 TB NVMe12 TB
nb-metal-32
128 vCPU256 GB2.4 TB NVMe20 TB
nb-metal-memMemory-Opt
32 vCPU512 GB960 GB NVMe12 TB
nb-metal-ioI/O-Opt
64 vCPU128 GB4.8 TB NVMe20 TB

All prices in USD. Egress above included bandwidth billed at $0.01/GB · IPv4 included · IPv6 included

Architecture

Silicon to socket,
nothing in between.

Conventional virtual machines insert a hypervisor between your code and the CPU. Nubis bare-metal removes that entire layer. Your OS boots directly on the physical machine — giving you raw clock cycles, direct NUMA topology access, and hardware performance counters that virtualisation typically masks.

  • Direct PCIe bus access for NVMe & GPU passthrough
  • NUMA-aware memory allocation without hypervisor interference
  • Hardware performance counters fully exposed (perf, eBPF)
  • SR-IOV networking — no virtio overhead
nubis-hardware-topology
$ nubis describe nb-metal-4


┌─────────────────────────────────────────────────────┐
│                   YOUR APPLICATION                  │
├─────────────────────────────────────────────────────┤
│               OPERATING SYSTEM KERNEL               │
├──────────────┬──────────────┬───────────────────────┤
│  CPU CORES   │  ECC DDR5    │   NVMe Gen5 Storage   │
│  AMD EPYC    │  UP TO 512GB │   UP TO 4.8TB RAW     │
│  UP TO 128C  │  4800 MT/s   │   PCIE 5.0 DIRECT     │
├──────────────┴──────────────┴───────────────────────┤
│           100 GbE DUAL-PORT NETWORKING               │
│        SR-IOV  ·  DPDK  ·  PRIVATE VPC              │
└─────────────────────────────────────────────────────┘

    ↑ NO HYPERVISOR   ↑ NO VIRTUAL DRIVER
    ↑ NO NOISY NEIGHBOR   ↑ NO CPU STEAL

Use cases

Built for workloads that demand the most.

High-Throughput Databases

Run PostgreSQL, MySQL, or MongoDB on dedicated hardware. Eliminate the I/O jitter that virtual disks introduce and achieve consistent sub-millisecond query latency.

Real-Time Analytics

ClickHouse, Apache Druid, and Spark workloads demand deterministic CPU scheduling. Bare-metal removes the noisy-neighbor risk and delivers predictable query performance.

CI/CD Build Farms

Full core allocation means your build pipeline actually runs at the clock speed advertised. No CPU steal, no throttling — just fast, parallel compilation.

Game Servers & Low Latency APIs

Single-digit millisecond response times at the application layer require zero interference from the infrastructure. Bare-metal delivers that guarantee.

Regulated Workloads

Single-tenant hardware provides a clean isolation boundary for PCI-DSS, HIPAA, and NDPR-regulated data — no shared kernel, no shared memory bus.

ML Training & Inference

Memory-optimised instances with up to 512 GB ECC DDR5 and high-bandwidth NVMe tiers let you load large model weights without swap pressure.

Zero latency.
Zero lock-in.

Reclaim your infrastructure. Deploy to our Lagos edge in under 60 seconds and experience what cloud performance actually feels like.

Simple pricing - No lock-in