How to Choose the Right netcup Server: An Engineering Sizing Guide
- Workload Divide: Use VPS G12.5 for burstable web/API traffic; use Root Server (RS G12.5) for sustained load, databases, and multiplayer game servers requiring zero CPU steal
- Memory Survival Line: Never run production MySQL 8.0 on 1 GB or 2 GB RAM—traffic spikes trigger the Linux OOM Killer; 4 GB RAM is the minimum production baseline
- Storage Economics: Don't upgrade your entire server tier just for disk space; attach Local Block Storage at €0.012/GB/month to cut storage costs by up to 60%
- Data Center Selection: Nuremberg (Germany) is netcup's owned flagship facility with 4–6 ms latency to DE-CIX Frankfurt; verify latency via Looking Glass before picking non-EU locations
- Contract Strategy: Lock in 24-month terms for 26% savings on stable production systems (and set a 31-day cancellation reminder), or use 1-month/Hourly billing for short-term testing
Table of Contents11 sections
The most expensive mistake in server provisioning is rarely a €5 difference on your monthly invoice—it is under-sizing critical resources, suffering mid-day database crashes from the Linux OOM Killer, and spending a sleepless weekend migrating production data.
netcup's catalog spans everything from €1 micro-slices to AMD EPYC Zen 5 monsters with dedicated physical cores and managed Plesk web hosting.
Skipping generic marketing claims, this guide breaks down server selection across five concrete engineering dimensions: CPU scheduling behavior, Linux memory & database survival boundaries, block storage economics, data center peering, and contract exit strategy.
Dimension 1: Workload Profile (Shared VPS vCores vs. Dedicated Root Server Cores)#
This is the primary fork in the netcup decision tree:
1. Burstable Workloads → Choose VPS G12.5#
- Typical Use Cases: Personal blogs, corporate websites, REST/GraphQL APIs, staging environments, cron automation, and lightweight Docker stacks.
- Physical Behavior:
- For more than 90% of a 24-hour cycle, baseline CPU utilization hovers between 5% and 20%;
- Compute spikes to 100% for only a few seconds or minutes during page renders, cache rebuilds, or package upgrades.
- Verdict: Shared vCores on VPS G12.5 are purpose-built for burstable traffic. You get strong single-core burst performance (1,800+ in Geekbench 6) and generous DDR5 ECC RAM at a fraction of dedicated compute cost.
2. Sustained High-Load & Zero-Jitter Workloads → Must Choose Root Server (RS)#
- Typical Use Cases: Multiplayer game servers (Minecraft, Palworld, CS2, Rust), primary transactional databases (MySQL/PostgreSQL), continuous video transcoding, CI/CD build farms, and Proxmox VE nested virtualization hosts.
- Physical Behavior:
- Game servers rely on deterministic single-threaded tick loops (e.g., 20 ticks per second). Even millisecond-level hypervisor scheduling jitter (CPU steal) on a shared VPS causes rubber-banding and tick drops for all connected players;
- Sustained compute pinning CPU cores at 100% for hours on a shared VPS will trigger fair-share hypervisor throttling.
- Verdict: Go straight to a Root Server (RS G12.5). With dedicated physical AMD EPYC 9645 (Zen 5) cores, near-zero CPU steal, and local NVMe arrays delivering 100,000+ IOPS, RS plans provide deterministic hardware performance.
Dimension 2: RAM Capacity & Surviving the Linux OOM Killer#
A classic pitfall among budget-conscious developers is ordering a 1 GB or 2 GB entry-level VPS, installing a full control panel alongside MySQL 8.0 and WordPress, and wondering why the site randomly throws 502 Bad Gateway or Error establishing a database connection under light traffic.
Why Modern Databases Crash on 1–2 GB RAM:#
- Baseline Database Memory Footprint:
On startup alone, a default MySQL 8.0 instance allocates 400 MB to 600 MB of resident physical memory for its InnoDB Buffer Pool, performance schema, and thread caches.
- Worker Process Multiplication:
In a typical PHP-FPM or Node.js/Python web stack, each concurrent worker consumes 30 MB to 60 MB of RAM. A burst of just 5–10 simultaneous requests can exhaust remaining memory in seconds.
- The Linux Kernel OOM Killer:
Once physical RAM and Swap are exhausted, the Linux kernel invokes the Out-Of-Memory (OOM) Killer to prevent a total kernel panic—automatically targeting and executing SIGKILL against the process with the highest memory score, which is almost always mysqld or postgres!
Safe Workload Boundaries by RAM Tier:#
| RAM Tier | Recommended Models | Safe Production Workload Boundary | What NOT to Run Here |
|---|---|---|---|
| 1 GB – 2 GB | VPS pico / nano | Monitoring agents (Uptime Kuma/Prometheus node), authoritative/recursive DNS (CoreDNS/AdGuard), SQLite apps, static Caddy/Nginx sites, single-binary Go/Rust services. | Do NOT run MySQL 8.0, GitLab, heavy JVM stacks, or bloated control panels. |
| 4 GB | VPS 500 G12.5 RS 500 G12.5 | Standard Production Baseline. Comfortably runs LEMP/LAMP, single-site WordPress/Ghost, Node.js/Django apps, and 5–10 lightweight Docker containers. | Avoid running multiple heavy Java/Spring Boot services simultaneously. |
| 8 GB | VPS 1000 G12.5 RS 1000 G12.5 | The Sweet Spot for Teams. WooCommerce/e-commerce stores, multi-site hosting, decoupled app + database stacks, and small CI/CD runners. | Handles the vast majority of small-to-medium production workloads. |
| 16 GB – 32 GB | VPS 2000 / 4000 RS 2000 / 4000 | Microservice clusters, high-concurrency relational databases, modded multiplayer game servers, self-hosted GitLab CE. | - |
Dimension 3: Storage Scaling Economics (Don't Upgrade Your Tier Just for Disk)#
In real-world operations, many servers sit below 30% CPU and RAM utilization while disk usage climbs past 85% due to user uploads, container images, logs, or local backups.
The Expensive Mistake:#
To expand storage from 128 GB to 512 GB, many users blindly upgrade their instance from a VPS 1000 (~€10.35/mo) to a VPS 4000 (~€32/mo)—overpaying by more than €21 every month for CPU and RAM they never touch!
The Smart Architecture: Attach Local Block Storage#
- netcup offers Local Block Storage volumes priced at just €0.012 per GB per month (net).
- Suppose you need an extra 300 GB of storage:
- Keep your existing VPS 1000 (€10.35/mo);
- Attach a 300 GB Local Block Storage volume:
300 × €0.012 = €3.60/mo; - Your total monthly bill is just €13.95/mo—saving over €18/month (more than 55% cost reduction) compared to upgrading the server tier!
- Local Block Storage is backed by a Ceph-based high-availability storage cluster and can be resized on the fly in the Server Control Panel (SCP) up to 8 TB.
Dimension 4: Data Center Location & Peering Rules#
Choose your data center based on where your users and upstream services actually live:
- Default Flagship: Nuremberg, Germany (
NUE):- Nuremberg is netcup's company-owned flagship facility with the lowest base pricing, fastest hardware refreshes, and largest inventory;
- Direct 4–6 ms fiber path to DE-CIX Frankfurt guarantees sub-20 ms latency across Central/Western Europe and ~30–35 ms to the UK.
- Regional Edge Nodes (
VIE,AMS,IAD,SIN):- Vienna (
VIE) is ideal for Austria, Switzerland, and Central/Southeastern Europe; - Amsterdam (
AMS) offers prime connectivity for the Benelux region, the UK, and transatlantic cables; - Manassas, Virginia (
IAD) serves North American audiences right inside the US East Ashburn/Northern Virginia corridor; - Singapore (
SIN) serves Southeast Asia—always test routing via our Looking Glass & Speed Test first to ensure your local ISP peers directly in Singapore rather than trombone-routing via the US or Europe.
- Vienna (
Dimension 5: Contract Term & Exit Strategy#
- Short-Term Testing, PoCs & Disposable Labs:
- Select a 1-month contract term or an Hourly billed tier;
- Spin up, benchmark, and cancel or delete when finished without locking yourself into a multi-year commitment.
- Core Long-Term Production Infrastructure:
- Choose the 24-month contract term to save roughly 26% off the monthly rate;
- Stack an official voucher code from our Coupons Hub at checkout for free months or instant discounts;
- Golden Operational Rule: As soon as your server is provisioned, either set a calendar reminder 45 days before contract expiration or immediately schedule cancellation at the end of the contract term (
zum Ende der Vertragslaufzeit) inside the CCP so you never accidentally trigger a 12- or 24-month auto-renewal under German contract law.