Guide
An archive of benchmark results for VPS servers that people choose to submit. This page explains what a report contains, how to read the numbers, and how far to trust them.
What a report contains
- The Summary tab: data parsed from the script output, split by script part (system info, disk, speed test, latency and so on), mostly as tables.
- The Raw output tab: the full pasted output, after redaction. The summary numbers come from it.
- The provider and location names and the optional plan note the submitter typed (shown as typed, not checked; reports are filed under a provider, city and data center, which may be empty), the script name and version, the test time recorded by the script, and the submission time.
- If a script is not recognised, or only partly: the unrecognised part shows raw output only, and the status says “partially parsed” or “raw only”.
What the L numbers mean
Many summary numbers carry a small tag such as L42: the number came from line 42 of the raw output. Click it to open the Raw output tab with that line highlighted and check that the parser read it correctly. Line numbers refer to the stored (redacted) text; redaction never adds or removes lines.
Supported scripts
Nine scripts are supported: eight are parsed into a summary, and TCPQuality is stored as raw text only. To contribute, copy a script's command from the submit page.
| Script | What it measures | Handling | Notes |
|---|---|---|---|
| YABS | System info, fio disk speed (4k / 64k / 512k / 1m), iperf3 bandwidth to public nodes, Geekbench 5 or 6 scores. | parsed | The most common script. -i and -g skip parts (missing from the report); -r shortens the iperf3 test. |
| bench.sh | System info, dd sequential write speed (3 runs), Speedtest against a few nodes. | parsed | Takes no options. A dd number is not an fio number. |
| HardwareQuality | Hardware review: CPU (including Geekbench 5), memory read/write, disk fio, and an overall score. | parsed | Runs Geekbench 5 only. |
| IPQuality | IP basics and type, risk scores from several databases, streaming / AI service unlock, mail port and blocklists. | parsed | Risk scores come from third-party databases; they say nothing about VPS performance. |
| NetQuality | Return routes to China's three carriers, latency to 31 provinces on each carrier, international iperf3 / speed tests. | parsed | The main source of China latency. One output can hold both an IPv4 and an IPv6 run. |
| NodeQuality | Hardware, IP quality and network quality tests in one output; each part is parsed on its own. | parsed | Each part means the same as in HardwareQuality / IPQuality / NetQuality. |
| 融合怪ECS | System info, CPU / memory / disk tests, streaming and IP quality, return routes, speed tests. | parsed | The most complex format; “partly parsed” is common. Check the raw output for anything not shown. |
| nws.sh | System and network info, Speedtest by region, plus optional iperf3 and route tests. | parsed | Shows only the items chosen at run time. |
| TCPQuality | TCP quality to carrier nodes in each province. | raw only | Stored as raw text with no summary; parsing may come once there are enough samples. |
Text from any other script can still be published; it appears as a raw report with no summary.
How to read the numbers
Only numbers from the same script and the same version compare directly; a few metrics have been checked and may be compared across scripts. By metric:
- Geekbench scores compare within a major version across scripts (for example Geekbench 5 from YABS and from HardwareQuality). Geekbench 5 and Geekbench 6 scores never compare: the two versions score on different scales. YABS may run either; HardwareQuality runs 5 only.
- China return latency compares only within one probe method: TCP (NetQuality, and the NetQuality part of NodeQuality, a large-packet TCP probe) and ICMP (ECS's ping) read differently and are never mixed; the three-carrier average uses the provinces that answered and shows how many. iperf3 bandwidth compares across scripts.
- Disk (fio), CPU (sysbench), memory and Speedtest compare only with results from the same script. Each script uses its own parameters (threads, duration, file size) and picks its own speed-test nodes.
- fio and dd do not compare. fio (YABS, HardwareQuality) tests random read/write by block size; dd (bench.sh) tests sequential writes only, is easily affected by caching, and forms its own group.
- YABS fio speeds (MB/s) split at v2026-07-03. That version changed how fio's KiB/s is converted to MB/s, so the same hardware reads about 2.4% higher; do not compare MB/s across that line. IOPS are not affected.
- Network numbers move with time and endpoint. Speed and latency depend on the time of day, test node and route. A latency where every probe failed is shown as
—. The signal bars next to latencies use <80 / <150 / <220 ms steps, only to make scanning easier. - One run is not long-term behaviour. A report is one run by one person; a disk test on a RAM-backed filesystem can read far too high, and noisy neighbours change results.
How far to trust a report
- Every report is marked User-submitted, unverified. This site never logs in to your server, cannot confirm the output came from a real run, and cannot check that the provider, location and plan note are right.
- The summary is parsed by software, once, at submission. If the parser is upgraded later, existing reports are not re-parsed automatically. If a number looks wrong, trust the Raw output tab.
- If you spot faked or tampered data, or personal information, use the Flag button on the report page. Administrators review flags and can hide a report.