Netcup vs Contabo (2026): Dedicated EPYC Zen 5 vs High-Ratio Shared vCPU Oversubscription

When evaluating budget-friendly German infrastructure providers, comparing headline RAM capacity in isolation often leads to severe production sizing errors. A virtual machine advertised with 12 GB or 16 GB of RAM on a densely consolidated host node cannot utilize that memory effectively if its vCPUs are stalled in hypervisor run queues or its block storage controller is bottlenecked by dozens of noisy neighbors.

Because Contabo separates its portfolio into high-density shared "Cloud VPS" plans and significantly more expensive dedicated "Cloud VDS" tiers, while Netcup offers both shared-vCore "VPS G12.5" and low-cost dedicated-core "Root Server G12.5" instances, an accurate comparison must evaluate all four tiers side by side. The matrix below details verified checkout pricing (both 19% DE VAT gross and 0% VAT net), CPU scheduling models, network port shapers, and international location surcharges.

Why can two 8 GB RAM Linux servers running the exact same Docker Compose stack exhibit a 5x difference in p99 HTTP response latency? The answer lies in hypervisor consolidation ratios. When a budget provider packs dozens of high-RAM virtual machines onto a single dual-socket physical host, two hardware bottlenecks surface simultaneously: CPU scheduler run-queue saturation and storage controller queue depth exhaustion.

Inside the Linux kernel, when your virtual machine has runnable threads waiting in its run queue (r column in vmstat), the KVM hypervisor must schedule your vCPU thread onto a physical core. If the physical host is heavily oversubscribed, the hypervisor forces your vCPU to wait while other tenants execute—incrementing the %st (steal time) counter. Even worse, when 80+ neighboring VMs flush journaled filesystem writes (ext4/xfs) or swap memory to disk at the same time, your guest kernel stalls in uninterruptible sleep (D state), driving %wa (I/O wait) above 15%–30%.

Netcup solves this in two ways: first, its VPS line maintains conservative vCore-to-physical-core consolidation ratios backed by enterprise NVMe/SSD RAID pools; second, its Root Server line (starting at €10.86/mo net / €12.93/mo gross for RS 500 G12.5 and €18.26/mo net / €21.73/mo gross for RS 1000 G12.5) reserves dedicated AMD EPYC 9645 Zen 5 cores so your compute threads are isolated from neighbor CPU spikes.

# 1. Monitor run queue (r), blocked I/O tasks (b), I/O wait (wa), and CPU steal (st) over 10 seconds
vmstat 1 10

# 2. Inspect block device queue depth (aqu-sz) and average write await (w_await in ms)
# Healthy local NVMe RAID should show w_await < 1.5 ms; saturated shared hosts often spike > 25 ms
iostat -xz 1 5

procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 1  0      0 4812400 142300 2104500   0    0     4    18  210  415  3  1 96  0  0

Device            r/s     rkB/s   rrqm/s  %rrqm r_await rareq-sz     w/s     wkB/s   wrqm/s  %wrqm w_await wareq-sz  aqu-sz  %util
sda              1.20     14.40     0.00   0.00    0.21    12.00   14.80    182.40     2.10  12.43    0.34    12.32    0.01   0.52

Contabo advertises "32 TB of Traffic" across its Cloud VPS and Cloud VDS tiers—a number that sounds enormous until you inspect the virtual NIC traffic shaper (tc / hypervisor policer). On Contabo’s entry Cloud VPS 10, the port reaches a hard ceiling of 200 Mbit/s (~25 MB/s), and even its €39+/mo dedicated Cloud VDS S is capped at 250 Mbit/s (~31.25 MB/s).

In real-world DevOps workflows, a 200 Mbit/s port cap creates painful operational friction: pulling a 3 GB multi-stage Docker image, restoring an off-site PostgreSQL backup from S3/Backblaze B2, or serving concurrent video/asset downloads immediately saturates the 200 Mbit/s pipe, causing TCP retransmissions and elevated ping times (bufferbloat) for active SSH and HTTPS sessions.

Netcup provisions a full 2,500 Mbit/s (2.5 Gbit/s) link across both its G12.5 VPS (VPS 500–8000) and G12.5 Root Server lines (12.5x faster than Contabo VPS 10 and 10x faster than Contabo Cloud VDS S), and even its €1.85/mo entry VPS pico G11.5s includes a 1,000 Mbit/s (1 Gbit/s) link.

Beyond raw hypervisor scheduling and network port ceilings, the total cost and operational friction of running a multi-region server fleet depend heavily on checkout fee transparency and Day-2 hypervisor tooling. Many budget hosts advertise a low base rate that only applies to a single domestic availability zone with restricted storage options, adding recurring monthly fees during checkout for non-EU regions, full NVMe disk capacities, or custom ISO mounting.

At Netcup, the Server Control Panel (SCP) exposes enterprise KVM lifecycle features—including custom ISO uploads, HTML5 VNC rescue access, hypervisor-level stateful firewalling, and Copy-on-Write snapshots—at €0 extra cost across all plans. When auditing the complete checkout cart across global locations and Day-2 management utilities, pay close attention to these four operational differences:

• ▪ Transparent Regional Pricing at Netcup: Base prices apply to "No preference Europe (EU)" and standalone Nuremberg promotional voucher packages (€18.26/mo net / €21.73/mo gross on RS 1000 G12.5). If you manually pin a specific city, Manassas (Virginia, US East) costs the exact same as pinned Nuremberg or Vienna (€21.00/mo net on RS 1000 G12.5; €14.01/mo net on VPS 1000 G12.5), while Singapore carries a transparent APAC transit surcharge.
• ▪ Custom ISO & Bare-Metal Style Recovery: Netcup’s Server Control Panel (SCP) lets you mount any official or user-uploaded .iso image (Proxmox VE, NixOS, Alpine, FreeBSD, Windows) and install over HTML5 VNC at €0 extra cost.
• ▪ Hypervisor Stateful Firewall: Netcup G12/G12.5 includes a configurable stateful firewall built directly into the SCP hypervisor bridge, allowing you to drop unwanted traffic before it consumes guest CPU interrupts.
• ▪ Routed IPv6 /64 Subnet: Netcup routes a full /64 IPv6 block with full reverse DNS (PTR) control in SCP, whereas Contabo’s rDNS and IPv6 tooling through its legacy customer panel is significantly less flexible for mail and container networking.

If your primary requirement is dedicated, non-oversubscribed CPU cores, the price gap between Netcup and Contabo is massive. Contabo requires stepping all the way up to its Cloud VDS S (~€39.27/mo gross) to escape shared vCPUs. At Netcup, dedicated AMD EPYC 9645 Zen 5 cores begin at €10.86/mo net / €12.93/mo gross on the RS 500 G12.5 (2 dedicated cores, 4 GB DDR5 ECC) and €18.26/mo net / €21.73/mo gross on the RS 1000 G12.5 (4 dedicated cores, 8 GB DDR5 ECC)—dropping to €15.22/mo net / €18.11/mo gross effective in Year 1 when redeemed with a standalone 2-Months-Free voucher (?gutschein=<CODE>) from our live coupon tracker.

Even on the shared vCPU tier in Singapore or the United States, once you add Contabo’s non-EU location surcharge and NVMe option to a Cloud VPS 10/20, its monthly bill approaches or exceeds Netcup’s VPS 500 G12.5 (€6.94/mo net / €8.26/mo gross) or VPS 1000 G12.5 (€12.18/mo net / €14.50/mo gross with 1 Month Free)—while Netcup gives you 12.5x the network port speed (2.5 Gbit/s vs 200 Mbit/s).

Weighing Netcup against Contabo comes down to whether your workload requires passive memory allocation for an idle test sandbox, or deterministic CPU execution, low-latency disk commits, and multi-gigabit network throughput for production traffic. When host oversubscription and 200 Mbit/s port shapers are factored into the equation, Netcup’s G12.5 architecture delivers substantially lower tail latency per euro.

Use the following workload criteria to select the right tier across both providers:

• ▪ Choose Netcup RS 500 / RS 1000 G12.5 (€10.86–€18.26/mo net | €12.93–€21.73/mo gross + 2 Months Free on RS 1000) for: Any production application, SaaS backend, PostgreSQL/MySQL database, Mailcow server, Coolify PaaS host, or game server where consistent latency, zero CPU steal, and 2.5 Gbit/s throughput matter.
• ▪ Choose Netcup VPS 1000 / 2000 G12.5 (€12.18 / €22.62/mo net | €14.50 / €26.92/mo gross + 1 Month Free) for: Balanced shared-vCore Linux workloads, staging environments, and multi-container Docker stacks that benefit from 2.5 Gbit/s networking and local SSD RAID reliability.
• ▪ Choose Contabo Cloud VPS only if: You are running an offline/low-traffic development lab that strictly requires 16 GB–24 GB of passive RAM at the absolute lowest possible monthly price and is completely insensitive to 200 Mbit/s bandwidth caps or I/O jitter.