netcup Benchmark & Performance Verification Guide (YABS, Geekbench 6, fio, iperf3)

Table of Contents10 sections

When your newly provisioned netcup Root Server or VPS comes online, do not rush straight into deploying production containers.

Spending 15 minutes running a standardized benchmark suite verifies that your hardware matches official specifications, checks host node health (%st CPU steal time), and gives you an objective baseline report (benchmark-report.txt) for future troubleshooting or support tickets.

This guide walks through the standard open-source benchmarking workflow and provides verified netcup G12.5 hardware reference scores.

---

1. All-in-One Automated Benchmark: YABS (Yet Another Bench Script)#

Instead of outdated legacy scripts, the global hosting and sysadmin community relies on YABS (Yet Another Bench Script) as the gold standard for Linux server verification.

YABS automatically runs three industry-standard tests without cluttering your system:

1. fio (Flexible I/O Tester): Measures real-world mixed random read/write throughput and IOPS across 4k, 64k, 512k, and 1m block sizes.
2. iperf3 Multi-Location Network Test: Tests upload and download throughput against public endpoints in Europe, North America, and Asia.
3. Geekbench 6: Cross-platform single-core and multi-core CPU compute benchmark.

How to Run YABS#

Connect via SSH and execute:

# Run the full suite (fio + iperf3 + Geekbench 6)
curl -sL yabs.sh | bash

Tip for Low-RAM Instances: On ultra-compact instances like VPS pico (1 GB RAM) or when you want to skip the 5-minute Geekbench CPU test, pass the -g flag:
curl -sL yabs.sh | bash -s -- -g

---

2. CPU Compute Baselines (Geekbench 6 & CPU Steal Check)#

In the latest G12.5 generation, netcup upgraded its Root Server line to flagship AMD EPYC 9645 (Zen 5 architecture) processors, while the x86 VPS line runs on high-density AMD EPYC host nodes and ARM VPS plans run on Ampere Altra processors.

Verified Geekbench 6 Reference Scores (v6.3+)#

How to Check Host Health (%st CPU Steal Time)#

Run top in your terminal and inspect the st (Steal Time) metric at the end of the %Cpu(s) line:

• On a Root Server (RS), %st should remain at 0.0% because your KVM cores are dedicated.
• On a VPS, %st should stay between 0.0% and 1.0% under normal conditions. If %st stays above 10% for prolonged periods, a noisy neighbor may be abusing the host node—save your YABS output and open a support ticket.

---

3. Storage Benchmarking (fio): Focus on 4K Random IOPS#

Many hosting providers advertise multi-gigabyte sequential speeds (1M block size), which only reflect copying huge video files.

In real production workloads—PostgreSQL/MySQL transactions, Redis persistence, Docker image extraction, and Node.js/PHP file reads—over 95% of disk operations are 4K random reads and writes (4k block size).

Reading Your YABS fio Output#

Look specifically at the 4k row in the fio summary table:

-----------------------------------------------------------------------
Block Size    Read IOPS              Write IOPS
4k            112.5k (439.45 MB/s)   98.2k (383.59 MB/s)  <-- Focus on this row!
64k           32.1k  (2.05 GB/s)     28.4k (1.78 GB/s)
512k          8.4k   (4.31 GB/s)     7.9k  (4.05 GB/s)
1m            4.6k   (4.71 GB/s)     4.3k  (4.41 GB/s)

Note: With the G12.5 refresh, netcup standardized VPS x86 storage on redundant enterprise SSD RAID10 arrays to keep monthly prices ultra-low despite global flash price spikes, while keeping pure high-IOPS NVMe on the Root Server (RS) line.

---

4. Network Throughput Testing (iperf3 Multi-Stream Best Practices)#

When running a single-thread speed test to a server across the Atlantic or in Asia (80–200 ms RTT), you might see 250 Mbps instead of the full 1 Gbps or 2.5 Gbps port speed.

This is normal TCP physics (Bandwidth-Delay Product), not a throttled NIC:
A single TCP stream waits for ACK packets across long distances before expanding its congestion window.

How to Measure True Physical Port Capacity#

1. Always pass -P 8 (8 parallel streams) to iperf3:

   # Test outbound throughput with 8 parallel streams
   iperf3 -c speedtest.netcup.net -P 8

   # Test inbound (download) throughput with -R (reverse mode)
   iperf3 -c speedtest.netcup.net -P 8 -R
   

1. Expected Line-Rate Results:
   • Against regional European or North American endpoints, an 8-stream test consistently saturates 940+ Mbps on 1G VPS ports and 2.35+ Gbps on 2.5G Root Server ports.
   • To maximize single-stream speed across high-latency international paths, apply the TCP BBR + 64 MB socket buffer tuning in our Network Architecture & BBR Guide.

---

5. Archiving Your Baseline Hardware Report#

Save a copy of your initial YABS output inside /root/benchmark-baseline.txt:

curl -sL yabs.sh | bash | tee /root/benchmark-baseline.txt

If your application ever feels sluggish months later, running a fresh benchmark and comparing it side-by-side against /root/benchmark-baseline.txt immediately tells you whether disk IOPS, CPU steal, or network throughput has changed.