When you buy a new VPS, many people's first reaction is to quickly set up the environment and build a website. But before that, a crucial thing is often overlooked—verifying the machine's true performance.
The configurations advertised on vendor pages may seem impressive, but are the 4 cores truly 4 physical cores, or just severely oversold virtual cores? What are the actual read and write speeds of the NVMe SSD? How fast is the network connection to China? These questions can't be answered without running several benchmark scripts.
This tutorial will guide you step-by-step through using several mainstream benchmarking tools to comprehensively test a VPS's CPU, disk, network, and overall performance, ensuring you buy with confidence and use it with peace of mind.
Before benchmarking, understand why you need to test.
There's an open secret in the server market—overselling. A vendor might split a 32-core physical machine into 100 "2-core" virtual machines. Although each virtual machine is allocated 2 vCPUs, the actual usable physical CPU time is far less than expected. Only through benchmarking can the actual impact of overselling on performance be quantified.
Furthermore, the definition of a "core" varies drastically across different platforms—a single hyper-threaded core in an Intel Xeon E5 versus a single physical core in an AMD EPYC can result in a single-core performance difference of more than five times. Without benchmarking, you'll never truly know what you're buying.
The core significance of benchmarking lies in three points: verifying whether the vendor has delivered on their configuration promises, quantifying the performance gap between different solutions, and establishing a performance baseline for later monitoring of performance degradation.
Quick Start: bench.sh—Get a Quick Look at Your Machine in 5 Minutes
If you just want to quickly understand the basics of a new machine in a few minutes, bench.sh is the best choice.
bench.sh is the most classic one-click benchmarking script in the VPS community. A single command can test system configuration, hard drive read/write speeds, and network speeds to nodes worldwide. Its positioning is "compact, basic, and one-click"—no need to install a bunch of dependencies, and it clearly lists the core data after execution.
Run the command:
wget -qO- bench.sh |If wget is unavailable, curl can be used:
curl -Lso- bench.sh |What does bench.sh test?
It mainly outputs three parts:
- System information: CPU model/number of cores, whether AES-NI hardware encryption is supported, total disk space, memory, Swap, operating system version, TCP congestion control algorithm (showing whether BBR is enabled), virtualization type (KVM/OVZ, etc.)
- Disk I/O speed: Writes three times sequentially using the dd command and takes the average, simple and intuitive.
- Network speed test: Tests upload, download speeds, and latency to multiple nodes globally.
Limitations of bench.sh: It does not test CPU benchmarks (Geekbench). If you need more detailed CPU performance data, you need to use the following tools.
International Standard: YABS – A One-Stop Comprehensive Benchmark
If you need a standardized benchmark report that allows for comparison with VPS users worldwide, YABS (Yet Another Benchmark Script) is currently the most reliable choice. Maintained by developer masonr, YABS enjoys high activity on GitHub.
YABS offers comprehensive testing dimensions and standardized results. A single command performs three tasks sequentially:
1. Tests CPU single-core and multi-core performance using Geekbench 6.
2. Tests disk read and write speeds at different block sizes using fio.
3. Tests network upload and download speeds to multiple nodes globally using iperf3.
A complete run takes approximately 10 to 30 minutes.
Running command:
curl -sL yabs.sh |If the network is unavailable, use the alternative address:
wget -qO- https://raw.githubusercontent.com/masonr/yet-another-bench-script/master/yabs.sh |
How to interpret YABS results?
CPU score: Geekbench 6 is a cross-platform, authoritative benchmark tool. YABS directly calls its command-line version, allowing for comparison with scores from other VPS users worldwide.
| Geekbench 6 Single-Core | Score Performance Evaluation |
| < 400 Slightly weak | potentially severely oversold |
| 400 - 800 Average | typical for most entry-level VPS |
| 800 - 1200 Good | high-performance VPS |
| > 1200 Excellent | AMD Ryzen or high-frequency CPU |
Disk I/O Speed: YABS tests read/write speeds for 4K, 64K, 512K, and 1M block sizes to determine hard drive type:
Disk Speed Type Determination
< 200 MB/s HDD (Mechanical Hard Drive)
200 - 500 MB/s Standard SATA SSD
500 - 3000 MB/s NVMe SSD
> 3000 MB/s High-end NVMe RAID Array
Network Speed: Focus on test results from Asian nodes (e.g., Singapore, Tokyo), as this reflects the actual experience in China.
LemonBench: The Top Choice for Domestic Users – Clear Overview of Network Backhaul for All Three Networks
If your business primarily targets users in mainland China, LemonBench is an indispensable tool. Developed by domestic developers, it offers more comprehensive testing of the domestic network environment than YABS.
Run the command:
`wget -qO- https://raw.githubusercontent.com/LemonBench/LemonBench/main/LemonBench.sh | -s -- --fast`
LemonBench's core features:
- Three-network backhaul route detection: Automatically detects the backhaul path from the VPS to the backbone networks of China Telecom, China Unicom, and China Mobile, and identifies the line type—CN2 GIA, CN2 9929, or the ordinary 163 line, directly determining the latency and stability for domestic users.
- Domestic node download speed test: Tests the actual speed to multiple domestic nodes.
- Streaming media unlock detection: Tests whether the VPS can unlock mainstream services such as Netflix, Disney+.
For businesses targeting domestic users, the backhaul route is more important than the outbound route—the experience of accessing the server from within China is determined by the backhaul.
Professional-grade disk testing: fio—precisely assesses storage performance.
While YABS includes the fio test, if you want a more in-depth analysis of disk performance, you can run fio separately for a dedicated test.
Installing fio:
Ubuntu/Debian
apt-get update && apt-get install fio -y
CentOS/RHEL
yum install fio -y
4K Random Read/Write Test (Simulating Database Scenario):
fio --name=randread --ioengine=libaio --rw=randread --bs=4k --size=1G --numjobs=1 --runtime=60 --group_reporting
Mixed Read/Write Test (Simulating Real Business Load):
fio --name=test --size=1G --filename=testfile --bs=4k --rw=randrw --ioengine=libaio --iodepth=64 --runtime=60 --numjobs=4 --group_reporting
Key Metrics:
- IOPS (Input/Output Per Second): Higher is better; 4K random read/write is a key measure of database performance.
- Bandwidth: Speed of sequential read/write of large files.
- Latency: Read/write latency, lower is better.
In-depth CPU Testing: Geekbench and UnixBench
YABS has integrated Geekbench 6, but you can also run Geekbench separately or use the more traditional UnixBench.
Running Geekbench 6 independently:
wget https://cdn.geekbench.com/Geekbench-6.2.2-Linux.tar.gz
tar -xvzf Geekbench-6.2.2-Linux.tar.gz
cd Geekbench-6.2.2-Linux
./geekbench6
After the test, a result URL will be generated (e.g., `https://browser.geekbench.com/v6/cpu/1234567`). Open it to view detailed CPU performance analysis.
UnixBench is a more traditional CPU benchmark tool, containing 13 sub-tests (Dhrystone integer arithmetic, Whetstone floating-point arithmetic, File Copy file reading and writing, Pipe Throughput pipe communication, Process Creation process creation, etc.). Its tests are time-consuming, typically taking 30 minutes to several hours.
Installation and Running:
wget https://github.com/kdlucas/byte-unixbench/archive/v5.1.3.tar.gz
tar -xvf v5.1.3.tar.gz
cd byte-unixbench-5.1.3/UnixBench
Multi-core Test
./Run -c $(nproc)
After Benchmarking: How to Use the Results?
After running the benchmark and getting a bunch of numbers, how do you use them?
First, compare them to the reference values to determine the performance level. Based on the score range given above, determine the performance level of the machine's CPU, disk, and network. If a single-core Geekbench score is below 400, it indicates a potential serious overselling issue; if disk 4K random read/write IOPS is below 20K, running database applications may be difficult.
Second, compare with other products in the same price range. YABS benchmark results can be compared with those of other VPS users globally to help you determine if the price is worthwhile.
Third, use benchmark results as a performance baseline. Save the initial benchmark results and run them again a few months later for comparison—a significant drop in score indicates potential server performance degradation.
Fourth, match benchmark results to your business needs. High single-core CPU scores are suitable for web servers and API services, while high multi-core scores are suitable for video processing and parallel computing; high disk IOPS are suitable for databases, and fast large file read/write speeds are suitable for static file storage.
Why choose Jtti cloud servers?
Benchmarking tools can help you verify a machine's true performance, but only if the machine itself can withstand the benchmark test.
Jtti cloud servers are designed with robust quality that can withstand benchmark tests in mind:
- High-performance hardware: Equipped with enterprise-grade high-performance CPUs, both single-core and multi-core performance have undergone rigorous testing to ensure no overselling or cost-cutting.
- Enterprise-grade NVMe SSD: Disk read and write speeds far exceed those of ordinary SATA SSDs, with stable 4K random read and write performance in real-world testing.
- Premium CN2 GIA lines: Core nodes in Hong Kong, the United States, and Singapore are connected to CN2 GIA optimized lines, with clear backhaul routes for all three major networks.
- Dedicated bandwidth: All models come standard with dedicated bandwidth, ensuring stable and reliable network speed test results and eliminating the problem of "neighbors competing for bandwidth."
- Transparent and verifiable: Jtti provides users with a public Looking Glass test entry point, allowing verification of network quality before purchase.
If you are choosing a cloud server that can withstand benchmark tests, why not run the test script yourself to verify Jtti cloud servers? The data will speak for itself.
