Best VPS Benchmark Scripts to Test Server Speed (2026 version)

I have a strict habit: the moment I get a new VPS, I never rush into installing software. Instead, I run test scripts first. This habit has saved me multiple times — I’ve caught several cases of overselling where the advertised “NVMe” only delivered 120 MB/s IO, and the “dedicated CPU” scored barely half of what similar-priced machines normally achieve. Catching these issues early lets you request a replacement or refund during the trial period. Discover them later, and you’re stuck with it.

Bottom line first: These three things must be tested on any new VPS

Whether the CPU performance is real, whether the disk is genuine NVMe or just a regular SSD, and the actual quality of the network route — none of these can be trusted based on the provider’s marketing page alone. You have to run real tests.

Here’s the standard veteran checklist:

1. Run YABS for an overall performance snapshot
2. Run Speedtest to check real bandwidth
3. Run BestTrace / MTR to inspect the return route

These three steps are usually enough to decide whether the VPS is worth keeping.

YABS: The most popular all-in-one test

curl -sL yabs.sh | bash

YABS has become the de facto standard in the VPS community. It tests CPU (via Geekbench), disk IO (via fio), and network speed (via iperf3). The output is standardized, making it easy to compare results across different machines and providers.

When people share VPS reviews on forums, they almost always post their YABS results. If you only have time to run one script, make it YABS.

Pay special attention to these numbers:

• Geekbench single-core score: This directly impacts WordPress dynamic page speed and AI inference performance. Below 500 is basically trash, 500–1000 is average, 1000–1500 is decent, and 1500+ is what you expect from good high-frequency instances.
• Disk write speed: Under 100 MB/s usually means HDD or very poor SSD. 100–300 MB/s is typical for regular SSDs. Over 500 MB/s is what real NVMe should deliver. I’ve seen plenty of “NVMe” listings that only manage 200 MB/s in reality — very common on cheap VPS plans.

Geekbench: Pure CPU testing

curl -L -o gk5.sh https://rebrand.ly/gk5 && bash gk5.sh

If you plan to run AI inference, code compilation, or high-concurrency WordPress, single-core performance is king. Geekbench focuses exactly on that.

After running, you’ll get a score you can compare against other CPUs on the official Geekbench website to understand exactly what level your VPS CPU is at.

nench: Great for quick filtering

curl -s wget.racing/nench.sh | bash

This test finishes in about a minute and is perfect when you’re comparing multiple VPS quickly. If the disk IO already looks terrible in nench, there’s no need to waste time on deeper tests. It’s the most efficient way to weed out bad machines early.

Fusion Monster (LemonBench / ECS): The most comprehensive test

International version:

curl -fsL https://ilemonra.in/LemonBenchIntl | bash -s fast

Or the version more commonly used by Chinese users (ECS Fusion Monster):

bash <(wget -qO- --no-check-certificate https://gitlab.com/spiritysdx/za/-/raw/main/ecs.sh)

This script provides the richest information — CPU, disk, streaming unlock status, triple-network return routes, latency, and more. It’s basically a full “physical exam” for your VPS. It takes longer than YABS, but if you want to understand the machine thoroughly, this is the one worth running.

Network Testing: The step many people skip

Bandwidth and latency have the biggest impact on real user experience, yet many people stop after testing CPU and disk.

Install and run Speedtest:

apt install speedtest-cli -y
speedtest-cli

Check the three key numbers: download speed, upload speed, and latency. If a “100 Mbps” plan only delivers 20 Mbps in reality, you’re dealing with heavy overselling or a bad route.

For return route quality, use BestTrace or MTR:

# Install MTR
apt install mtr -y

# Test return route to China (Alibaba DNS)
mtr -r -c 100 223.5.5.5

In the output, look at the node IPs: - Starting with 59.43 → Telecom CN2 (good) - Starting with 202.97 → Regular Telecom 163 (average) - AS9929 → China Unicom premium line - AS4837 → Regular Unicom line These directly determine how fast users in China will actually experience your VPS.

Important tip: Always test again between 8 PM and 11 PM (China time). Many VPS look fine during the day but get heavily throttled during evening peak hours. Daytime results can be misleading.

Advanced Disk IO Testing (for serious workloads)

If you plan to run databases or any I/O-heavy applications, use fio for more accurate testing:

apt install fio -y

# Random 4K read/write test (best reflects real database performance)
fio --name=randtest \
    --filename=testfile \
    --size=1G \
    --bs=4k \
    --rw=randrw \
    --iodepth=64 \
    --runtime=30 \
    --time_based \
    --group_reporting

Pay attention to IOPS (operations per second) and bandwidth, not just sequential speeds. Many “NVMe” disks boast 2000+ MB/s sequential but deliver only tens of thousands of 4K IOPS — which feels mediocre in real database scenarios.

Performance Judgment Guide

Use-case priorities

Website building & SEO sites: Focus mainly on disk IO and single-core CPU — these two directly affect WordPress page response times.

Game servers: Latency and packet loss matter most, followed by CPU. A 10 ms difference in latency affects gameplay far more than doubling the CPU score.

AI tools and compilation: Prioritize Geekbench single-core and multi-core scores, then disk IO.

Dedicated IPs, streaming, or proxy nodes: Focus on streaming unlock status and network route quality. The Fusion Monster script is ideal here.

Practical advice

Before purchasing, many providers publish official test IPs. Ping them from your location and run MTR to check route quality first. After buying, use the refund window (often 30 days) to run full benchmarks. If something feels off, request a refund or replacement immediately — don’t wait until the trial period ends.

Providers like RackNerd, CloudCone, and DMIT all provide clear test IPs. Spending a few minutes testing before you buy is absolutely worth it — don’t skip this step.