You’ve just purchased a new VPS advertised as featuring "high-performance CPUs," "NVMe SSDs," and "premium CN2 GIA routing"—but how does it actually perform? Will speeds drop during evening peak hours? Is the disk I/O performance throttled? Is the CPU heavily oversold? You won't find the answers to these questions on the marketing page.

The only reliable method is to run your own performance tests. This guide walks you through a comprehensive VPS performance assessment—covering CPU, memory, storage, and network—using industry-standard tools. Whether you are a VPS novice or a seasoned user looking to systematically verify a machine, following these steps will yield a reliable performance report.

Pre-test preparation: Confirming the remote machine and environment

Many network testing tools require two machines working in tandem. We recommend using your VPS as the server and another machine you control (such as your home computer or a separate cloud server) as the client. This ensures you measure the actual throughput between the two points, eliminating interference from third-party nodes.

Before starting, update your system and install the necessary basic tools:

# Ubuntu / Debian

apt update && apt install -y curl wget sudo

# CentOS / Rocky Linux

yum update -y && yum install -y curl wget sudo

If the VPS is running production workloads, we recommend performing tests during off-peak hours to ensure testing traffic does not impact normal service operations.

CPU Testing: Core count doesn't equal performance

The CPU is the computational heart of a VPS, but "core count" refers only to quantity; true performance is determined by single-core efficiency and the processor's architectural generation. For instance, given the same core count, the single-core performance of an Intel Xeon Gold 6248R can be more than three times that of an older E5-2680.

Geekbench 6 is currently the most authoritative cross-platform CPU benchmark. It is already integrated into the YABS script, but it can also be run 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

Once the test is complete, an online results link will be generated, allowing you to view detailed single-core and multi-core scores. Interpreting the scores: A single-core score below 400 indicates a weak CPU, potentially suggesting severe over-provisioning; 400–800 is average; 800–1,200 is good; and scores above 1,200 are excellent.

UnixBench is a more traditional comprehensive benchmark comprising 13 sub-tests that cover integer arithmetic, floating-point operations, file copying, pipe communication, process creation, and more. You can run a multi-core test using the command `./Run -c $(nproc)`. Its scores allow for benchmarking against other VPS instances worldwide, making it suitable for assessing overall performance.

Memory Testing: Beyond Capacity, Look at Bandwidth

Memory testing is often overlooked, yet memory bandwidth and latency significantly impact databases and caching services. `sysbench` can be used to test memory read/write speeds:

# Installation

apt install sysbench -y

# Test memory write speed

sysbench memory --memory-block-size=1M --memory-total-size=10G run

The `transferred` and `throughput` values ​​in the output indicate memory bandwidth. The bandwidth difference between DDR4 and DDR5 can exceed twofold; if test results are significantly lower than the norm for that generation of memory, it may indicate resource contention.

`stream` is another classic memory bandwidth testing tool; it measures bandwidth for four specific operations—Copy, Scale, Add, and Triad—reflecting access patterns closer to real-world applications.

Disk Testing: IOPS Matters More Than Sequential Read/Write

Disk performance directly affects database queries, file I/O, and website response times. `fio` is the industry-standard tool for storage benchmarking, capable of simulating a wide range of real-world workloads. # Installation

apt install fio -y

# 4K random read/write test (simulating database scenarios)

fio --name=randrw --ioengine=libaio --rw=randrw --bs=4k \

--size=1G --numjobs=1 --runtime=60 --group_reporting

# Sequential read/write test (simulating large file transfers)

fio --name=seqrw --ioengine=libaio --rw=rw --bs=1M \

--size=1G --numjobs=1 --runtime=60 --group_reporting

Focus on IOPS (Input/Output Operations Per Second) and latency. 4K random read/write IOPS below 20K indicate weak disk performance; NVMe SSDs typically achieve over 50K. Regarding latency, read/write latency under 5ms is excellent, while latency exceeding 10ms can impact database performance.

If you prefer not to run fio manually, the YABS script automatically performs tests using four block sizes—4K, 64K, 512K, and 1M—and identifies the disk type.

Network Testing: Bandwidth, Latency, and Packet Loss

Network testing covers two dimensions: bandwidth throughput and latency/packet loss.

iperf3 is used to accurately measure actual bandwidth between two points. Start `iperf3 -s` on the server side, and execute the following on the client side:

# Upload test (Client → Server)

iperf3 -c Server_IP -t 60 -P 10

# Download test (Server → Client); add -R

iperf3 -c Server_IP -R -t 60 -P 10

If the measured bandwidth is significantly lower than the advertised rate, the bandwidth may be overstated or subject to shared oversubscription. A high number of retransmissions indicates severe packet loss on the link.

MTR is used to test latency and packet loss while displaying the routing path for each hop:

mtr -rwzc 100 Target_IP

For CN2 GIA connections from mainland China to the US West Coast, the normal average latency is between 130ms and 180ms; for CN2 GIA connections to Hong Kong, it is typically 30ms to 60ms. If the packet loss rate exceeds 5% during evening peak hours, the connection quality is substandard. One-Click Comprehensive Benchmarking: bench.sh and YABS

If you prefer not to run tools individually, two scripts offer an all-in-one solution for comprehensive testing.

bench.sh is ideal for a quick assessment. A single command outputs system information, disk I/O speeds, and network speeds to global nodes. Simply run `wget -qO- bench.sh | ` to get results within five minutes.

YABS is a more comprehensive option. It automatically utilizes Geekbench 6 for CPU testing, fio for disk testing, and iperf3 for network testing, with a full run taking approximately 10–30 minutes. Executing `curl -sL yabs.sh | ` generates a standardized performance report, allowing for direct side-by-side comparisons with benchmark scores from other VPS providers worldwide.

Post-Testing: Jtti Ensures Foundational Performance

Once testing is complete, you can gauge the true capabilities of a VPS. However, testing is merely a means to an end; the ultimate determinant of service quality is the server's underlying hardware.

Jtti’s cloud server solutions provide robust hardware-level support for high performance. The entire lineup comes standard with enterprise-grade NVMe SSDs, delivering 4K random read/write IOPS in fio tests that far exceed those of standard SATA SSDs. Nodes in Hong Kong and the US feature premium CN2 GIA connectivity with optimized direct routing across major carriers; MTR tests confirm that return traffic travels exclusively via the 59.43 dedicated network nodes, maintaining a packet loss rate below 0.5% even during peak evening hours. Dedicated bandwidth is standard across all plans, ensuring iperf3 results closely match the advertised specifications—eliminating speed test inaccuracies caused by "noisy neighbors" competing for bandwidth.

A "same-price renewal" policy guarantees that the renewal cost matches the initial purchase price, ensuring predictable long-term operating costs amidst the anticipated price hikes of 2026.