What Is It?
Compute Engine provides virtual machines with two server types: Dedicated Core servers with physical CPU cores and vCPU servers with cost-effective virtual CPUs. Both types offer customizable resources (CPU, RAM, storage, networking), support Linux and Windows operating systems, and feature Live Vertical Scaling to adjust resources without downtime. Servers run within Virtual Data Centers (VDCs) with availability zone support for high availability.
Quick Facts
| Aspect | Details |
|---|---|
| Server Types | Dedicated Core, vCPU servers, Cloud GPU VMs, Confidential VM (own page) |
| CPU Options | AMD EPYC or Intel Xeon, selectable on Dedicated Core only; the CPU models offered vary by location, and vCPU Servers are not available in DE/FKB or US/LAS |
| Scaling | Live Vertical Scaling (LVS) for CPU and RAM |
| Availability | Two compute availability zones (Zone 1, Zone 2) plus Auto; 99.95% monthly uptime SLA (Cloud GPU VMs are excluded from the SLA) |
| Storage | Essential, Balanced, Performance (formerly HDD, SSD Standard, SSD Premium); 1 GiB to 4 TiB per volume, larger on request |
| Management | Data Center Designer (DCD), Cloud API, SDKs, Terraform |
Choosing Your Server Type
| Aspect | Dedicated Core | vCPU servers |
|---|---|---|
| Resources | Physical CPU cores (hyper-threaded) | Virtual CPUs (shared) |
| CPU Selection | AMD or Intel, switchable after provisioning | No CPU family selection |
| Performance | Guaranteed, consistent | Variable based on host utilization |
| Best For | Databases, real-time processing, performance-intensive workloads | General workloads, dev/test, SaaS, web apps |
| Price Point | Higher: 0,036 to 0,046 EUR per core per hour (by CPU model) and 0,0045 EUR per GB RAM per hour | Lower: 0,012 EUR per vCPU per hour and 0,002 EUR per GB RAM per hour |
Choose Dedicated Core when: You need guaranteed CPU performance, consistent latency, or specific CPU architecture (AMD/Intel).
Choose vCPU when: Cost optimization is priority, workloads tolerate variable performance, or you need flexible scalability.
Dedicated Core Specifications
| Resource | Minimum | Maximum |
|---|---|---|
| Physical Cores | 1 core | 62 cores |
| RAM | 0.25 GB | 230 GB |
| PCI Connectors | 0 | 24 (NICs and storage combined) |
| CD-ROM Devices | 0 | 2 |
| Storage per Volume | 1 GiB | 4 TiB (larger on request) |
Note: Each physical core is hyper-threaded (2 logical CPUs) on AMD EPYC and Gen-5 Intel Xeon hosts (Haswell, Broadwell, Skylake, Ice Lake); Gen-6 Intel Xeon Sierra Forest hosts present each physical core as a single logical core with no hyper-threading.
vCPU Server Specifications
| Resource | Minimum | Maximum |
|---|---|---|
| vCPUs | 1 vCPU | 60 vCPUs |
| RAM | 0.25 GB | 230 GB |
| PCI Connectors | 0 | 24 (NICs and storage combined) |
| CD-ROM Devices | 0 | 2 |
| Storage per Volume | 1 GiB | 4 TiB (larger on request) |
What You Can Do
Live Vertical Scaling
Add CPU, RAM, NICs, or storage while the VM is running. Linux supports full LVS for both CPU and RAM. Windows supports live CPU scaling only, and scaling beyond eight CPU cores requires a reboot; RAM changes on Windows require a reboot. Above 240 GB of RAM, RAM hotplug is disabled and every RAM increase restarts the VM.
Flexible Resource Configuration
Choose CPU cores (or vCPUs), memory, and storage freely and independently on Dedicated Core and vCPU servers, and adjust them after creation via LVS. Cloud GPU VMs are the exception: they use fixed templates (H200-S, M, L, XL) whose CPU and RAM cannot be changed after provisioning.
Multi-Availability Zone Deployment
Spread servers and storage across separate physical availability zones within a region for fault tolerance. If one zone experiences issues, services in other zones remain operational.
Secure Virtualization
VMs are isolated through secure virtualization technology.
Core Technology Switching
For Dedicated Core servers, switch between AMD and Intel CPU architectures after provisioning. Requires VM restart but provides flexibility for workload optimization or specific CPU feature requirements.
Storage Flexibility
Attach Essential, Balanced, or Performance volumes (1 GiB to 4 TiB per volume, larger on request). Scale storage capacity up later without a VM restart. A server takes up to 24 volumes within its 24 PCI connector limit, of which at most 4 can be Performance.
Cloud GPU VMs
Deploy GPU-accelerated virtual machines powered by NVIDIA H200 GPUs for AI/ML training, inference, high-performance computing, and 3D rendering workloads. Cloud GPU VMs run only in Frankfurt (de/fra/2) and Worcester (gb/bhx) and come in four fixed templates (H200-S, M, L, XL with 1, 2, 4 or 8 GPUs) whose CPU and RAM cannot be scaled later. They support IONOS Linux images only, are limited by default to one H200-S per contract, and are not covered by the SLA. Cloud GPU VMs provide dedicated GPU resources with high-bandwidth GPU memory for demanding parallel compute tasks.
IPv6 Configuration
Enable IPv6 addressing for modern network requirements. Supports dual-stack configurations for compatibility with both IPv4 and IPv6 networks.
Best For
| Scenario | Server Type | Why |
|---|---|---|
| Production databases (PostgreSQL, MongoDB) | Dedicated Core | Guaranteed CPU performance and consistent I/O |
| Real-time analytics, HPC workloads | Dedicated Core | Predictable latency and dedicated resources |
| Web applications, API servers | vCPU | Cost-effective with adequate performance for variable load |
| Development and testing environments | vCPU | Lower cost, flexible scaling for temporary workloads |
| SaaS platforms, microservices | vCPU | Balance of performance and cost for distributed architectures |
| Latency-sensitive applications | Dedicated Core | Consistent performance without noisy neighbor effects |
| AI/ML training and inference | Cloud GPU VMs | Dedicated NVIDIA H200 GPUs for parallel compute workloads |
| 3D rendering and HPC | Cloud GPU VMs | High-bandwidth GPU memory for graphics and scientific computing |
Consider Alternatives If
| If You Need... | Consider | Why |
|---|---|---|
| Pre-configured, template-based VMs | Cubes | Quick deployment with suspend/resume capabilities |
| Containerized workload orchestration | Managed Kubernetes | Automated container management and scaling |
Key Considerations
Billing & Costs
- Main billing: Per hour based on cores/vCPUs and RAM allocated
- Dedicated Core: Premium pricing for guaranteed physical cores; 1-year or 3-year Cloud Savings Plans lower the hourly rate for cores and RAM
- vCPU: 0,012 EUR per vCPU per hour and 0,002 EUR per GB RAM per hour, against 0,036 to 0,046 EUR per Dedicated Core and 0,0045 EUR per GB RAM
- When stopped: Stopping the server in the DCD deallocates its CPU and RAM and suspends their billing; attached storage remains provisioned. Shutting down only inside the operating system keeps the resources allocated and billed.
- Additional costs: Storage (Essential, Balanced, Performance), outgoing data transfer, reserved public IPs, snapshots, and licences such as Windows Server, Microsoft SQL Server or RHEL (billed per vCPU or core)
Limitations
- Dedicated Core: Maximum 62 physical cores, 230 GB RAM per instance
- vCPU: Maximum 60 vCPUs, 230 GB RAM per instance
- PCI devices: 24 maximum (NICs and storage volumes combined), plus up to 2 CD-ROM drives
- Windows LVS: RAM scaling requires a reboot; CPU scaling is live up to eight cores, and scaling beyond eight cores requires a reboot
- CPU switching: Dedicated Core servers must restart when changing AMD/Intel architecture
- vCPU CPU family: Cannot select or change CPU family after creation
- RAM limits: Larger sizes available via sales contact, not standard self-service
Management Options
- Data Center Designer (DCD): Drag-and-drop visual interface for VM creation, configuration, monitoring, power management
- Cloud API: Full programmatic access to server lifecycle and configuration
- Compute SDKs: Language-specific libraries (Python, Go, Java, etc.)
- Terraform provider: Infrastructure-as-code for reproducible deployments
- Configuration management: Integration with Ansible, Puppet for automated setup