Netcup VPS vs Root Server (2026): KVM vCore Scheduling, Zen 5 Pinning & G12.5 Tier Math

In German hosting nomenclature, the term "Root Server" historically referred to bare-metal hardware, which often confuses international systems engineers when they inspect Netcup’s catalog. At Netcup, a Root Server (RS G12.5) is a hardware-virtualized KVM instance with dedicated, non-oversubscribed AMD EPYC 9645 ("Turin", Zen 5) processor cores and dedicated DDR5 ECC Registered memory, whereas a Netcup VPS (VPS G12.5 and G11.5s) is a KVM instance backed by shared, time-sliced AMD EPYC vCores.

Because both product lines execute inside the exact same qemu-system-x86_64 KVM hypervisor stack and are orchestrated via the same Server Control Panel (SCP) REST API, you do not sacrifice kernel isolation, custom ISO booting, eBPF/WireGuard support, or stateful edge firewalling when choosing a VPS. Instead, the architectural divergence lies in how the host kernel schedules your vCPU threads under load, the underlying storage media (PCIe NVMe vs enterprise SSD RAID), and the formal 99.9% annual availability SLA attached to Root Servers.

The matrix below contrasts the hypervisor scheduling, memory hierarchy, block storage subsystem, and network port shaping policies across both product families:

In Generation 12.5, Netcup offers a complete progression from ultra-low-cost G11.5s micro instances (VPS pico G11.5s at €1.85/mo net / €2.21/mo gross and VPS nano G11.5s at €3.10/mo net / €3.69/mo gross) up to 16-core / 64 GB RAM G12.5 servers. On the Root Server side, dedicated Zen 5 cores begin at RS 500 G12.5 (2 dedicated cores, 4 GB DDR5 ECC, €10.86/mo net / €12.93/mo gross) and reach their sweet spot at RS 1000 G12.5 (4 dedicated cores, 8 GB DDR5 ECC, €18.26/mo net / €21.73/mo gross, with a 2-Months-Free standalone voucher).

What makes the G12.5 pricing curve particularly interesting for capacity planners is that the relative price premium of a Root Server over an equivalently sized VPS is not constant across tiers. At the 8 GB RAM tier (1000 G12.5), the Year-1 effective gap between VPS 1000 G12.5 (€11.17/mo net) and RS 1000 G12.5 (€15.22/mo net) is only €4.05/month net—making the upgrade to 4 dedicated Zen 5 cores and PCIe NVMe storage an easy engineering decision. At the 32 GB (4000 G12.5) and 64 GB (8000 G12.5) tiers, however, reserving 12 to 16 dedicated physical Zen 5 cores nearly doubles the monthly rate relative to the shared-vCore VPS equivalent.

Examine the complete tier-by-tier G12.5 Net (0% VAT) and Gross (19% VAT) pricing, voucher math, and unit economics table below before choosing your server size:

Both product lines execute inside a Linux KVM hypervisor (qemu-system-x86_64), meaning neither suffers from the shared-kernel security surface or sysctl / kernel module restrictions of LXC or OpenVZ containers. Every VPS and Root Server boots its own independent Linux kernel, loads custom nftables, wireguard, or overlayfs modules, and manages its own TCP congestion control stack.

Beneath the guest kernel, however, the host hypervisor’s cgroup v2 CPU controller (cpu.max), scheduler run-queue isolation, and NUMA cache affinity policies behave differently depending on whether you provision a Root Server or a VPS:

# 1. Verify hypervisor vendor signature and hardware virtualization extensions (svm for AMD-V)
lscpu | grep -E "Hypervisor|Virtualization|Model name"

# 2. Measure kernel wake-up scheduling jitter (microseconds) using rt-tests cyclictest
# Install: apt-get update && apt-get install -y rt-tests
cyclictest --smp -p 80 -m -i 1000 -l 1000 -q

Hypervisor vendor:        KVM
Virtualization type:      full
T: 0 ( 4120) P:80 I:1000 C:   1000 Min:      4 Act:    8 Avg:    9 Max:      42
T: 1 ( 4121) P:80 I:1000 C:   1000 Min:      4 Act:    9 Avg:   10 Max:      48

Storage subsystem architecture is the second major differentiator between the two product lines. Netcup Root Server G12.5 nodes connect guest virtual machines via multi-queue virtio-scsi to local enterprise PCIe NVMe flash arrays backed by DDR5 ECC Registered memory. When a relational database executes heavy WAL checkpointing or fdatasync() commits at high queue depth, PCIe NVMe non-blocking queues maintain sub-200-microsecond O_DIRECT response times.

Netcup’s G12.5 VPS lineup (VPS 500 to VPS 8000) utilizes enterprise SSD RAID storage (while both lines support attaching up to 8 TB of additional Local Block Storage). For sequential container layer reads, static asset serving, and moderate transactional logging, enterprise SSD arrays easily saturate the virtual block interface, though heavy concurrent random 4K write bursts show higher tail latency than the Root Server’s PCIe NVMe pool.

On the networking side, Generation 12.5 unified the virtual NIC link speed across both families: every VPS 500–8000 G12.5 and every Root Server G12.5 connects via a 2.5 Gbit/s (virtio-net) interface, differing only in their 24-hour rolling traffic flatrate shaping thresholds:

• ▪ Dedicated Zen 5 Memory & PCIe NVMe Subsystem vs Enterprise SSD: Root Servers pair DDR5 ECC Registered memory with local enterprise PCIe NVMe flash arrays, delivering higher queue-depth random IOPS and lower fsync latency during heavy database checkpointing or large git/node_modules extractions, plus a 99.9% annual availability SLA and 30-day satisfaction guarantee. G12.5 VPS plans (VPS 500 to VPS 8000) use enterprise SSD storage (expandable up to 8 TB via Local Block Storage, while entry G11.5s pico/nano include 30 GB / 60 GB SSD), easily saturating sequential reads/writes for web serving and container startups.
• ▪ 2.5 Gbit/s Interface Across All G12.5 Tiers: Both Root Server G12.5 and VPS 500–8000 G12.5 instances link at 2,500 Mbit/s (~312 MB/s real-world TCP throughput) with an unmetered traffic flatrate (Root Server G12.5 shapes to 300 Mbit/s if rolling 24h traffic exceeds 3 TB; VPS 500–8000 G12.5 shapes to 200 Mbit/s if rolling 24h traffic exceeds 2 TB), while entry VPS pico/nano G11.5s instances link at 1,000 Mbit/s (~125 MB/s).

When designing a multi-service topology on Netcup, the most cost-effective approach is matching each node to its primary hardware bottleneck: CPU instructions per second and disk fsync IOPS versus resident memory footprint. Rather than overpaying for dedicated cores on a 32 GB Elasticsearch cache node or starving a production PostgreSQL primary on shared vCores, you can mix Root Servers and VPS nodes within the same data center location (such as Nuremberg NUE1 or Vienna VIE1) and link them over a private VLAN or WireGuard mesh.

Use the engineering matrix below to map your exact software stack to the optimal Netcup tier, and remember that you can scale up in the Customer Control Panel (CCP) or migrate live block volumes across the local 2.5 Gbit/s fabric as your workload grows: