RAW POWER.
ZERO OVERHEAD.
Nubis Bare Metal Instances provide dedicated physical servers with no virtualization layer. Experience maximum performance, consistent latency, and full hardware access for your most demanding workloads.
Why Bare Metal?
Traditional virtualization adds layers of complexity and latency. Nubis Bare Metal removes the hypervisor, giving you direct access to the silicon.
Latency
Performance
Overhead
Scalability
Dedicated Processors
Full access to AMD EPYC™ 9004 Series or Intel® Xeon® Scalable processors. No noisy neighbors.
- Up to 128 Cores / 256 Threads
- Base clock up to 3.1GHz
- AVX-512 Support
ECC RDIMM RAM
High-density DDR5 memory with error correction for mission-critical reliability.
- Up to 2TB DDR5 RAM
- 4800MT/s transfer speed
- Eight-channel memory architecture
NVMe Gen5 Storage
Direct-attached PCIe 5.0 NVMe drives for extreme I/O performance and low latency.
- Up to 30TB Raw NVMe
- Software RAID 0/1/10 support
- Sub-100µs latency
100Gbps Connectivity
Dual 100GbE ports with support for SR-IOV and high-performance packet processing.
- Redundant 100G uplinks
- DPDK & eBPF optimized
- Private VPC networking
Optimized for Scale
Nubis Bare Metal is engineered for workloads that demand 100% of the underlying hardware resources without compromises.
Large-Scale Databases
Eliminate hypervisor latency for mission-critical PostgreSQL, MongoDB, or Cassandra clusters with direct-path NVMe I/O.
Machine Learning & AI
Scale high-throughput model training and inference workloads with dedicated GPU access and massive memory bandwidth.
Financial Modeling
Deterministic performance for high-frequency trading and risk analysis where every microsecond of latency counts.
Custom Hypervisors
Run your own virtualization stack (KVM, Xen, ESXi) or container orchestration without nested virtualization penalties.
Global CDN Edge
Deploy custom caching engines and edge compute logic on the same hardware that powers our global network backbone.
Ad-Tech Realtime Bidding
Handle millions of requests per second with consistent sub-millisecond response times across millions of cores.