Sophia Stack X

Sophia Stack X benchmarks

Sophia Stack X benchmarks

These are numbers we measured on 15 November 2026 on one development computer. It is one machine, not a lab: the load tool ran on the same computer as the server and shared its CPU, nothing was tuned, and a different machine will give different results. We did not measure WordPress, Joomla or Ghost, so there are no comparisons here. Use the numbers to get a feel for the size of the program, not as a promise.

The machine and the setup

  • Windows 10 Pro, AMD Ryzen 7 1800X (8 cores, 16 threads), about 80 GB of RAM, Samsung 970 EVO Plus NVMe drive.
  • Program: the Windows release build of version 0.1.0 (Rust, optimized, thin LTO, stripped), 32,214,528 bytes.
  • Data: a copy of the site you are reading, 24 pages and about 130 MB of site files including downloads, in a data folder of its own.
  • Server on 127.0.0.1:8874, one process, no reverse proxy, no HTTPS, default settings.
  • Load tool: a small Python script using asyncio and raw keep-alive HTTP/1.1 connections, run on the same computer. A script like this can be the slowest part of the test, so treat the published-page throughput as a floor for the server, not a ceiling.

Results

WhatResult
Program size30.7 MiB (32,214,528 bytes), one file, nothing else to install
Cold start to first answer521 ms median of 7 starts (range 516 to 577 ms), measured from launching the process to the first 200 for a published page
Idle memory13.5 MiB working set, 10 seconds after start, before any request
Memory under load24 MiB peak working set with 32 connections on a published page, 28 MiB peak with 32 connections on the preview
Published page, 1 connection540 requests per second, p50 1.8 ms, p99 3.6 ms
Published page, 32 connections3,984 requests per second, p50 5.1 ms, p99 54 ms, no errors in 39,942 requests
Dynamic preview, 1 connection156 requests per second, p50 6.1 ms, p99 10.7 ms
Dynamic preview, 32 connections177 requests per second, p50 165 ms, p99 427 ms, no errors in 1,804 requests
Publish the docs site (24 pages)6.5 to 7.5 seconds, median 6.9 s, 5 runs

Each load run lasted 10 seconds, over the same page, /docs/concepts.

What the numbers mean

The published page is the path visitors normally use. It is a plain file served from a release folder, so one process answered several thousand requests a second with a few megabytes of extra memory. The p99 at 32 connections is probably the load script's own scheduling, since both ran on the same CPU.

The dynamic preview is the admin's page preview. It renders the page through the template on each request with drafts included, behind a signed-in session. It is meant for one editor, not for visitors, and the numbers show it. It is far slower than the published page because it does the work that publishing does once.

Cold start is the whole time from launching the process to the first answer. Publish time is the whole command, started from a new process, and it includes rendering every route, checking links, picture descriptions and placeholder text, and writing a new numbered release. The release for this site was about 8 MB on disk.

The preview run signed in through the install's one-click demo mode, so the session check had that extra middleware in its path.

How to reproduce

  1. Download the program from Install and verify it.
  2. Copy a data folder to a scratch place. Use sophia backup or copy it while the program is stopped.
  3. Time a cold start: launch sophia serve --data <folder> --bind 127.0.0.1:8874 and poll a page until it answers 200.
  4. Time a publish: sophia --data <folder> publish <site> several times, taking the median.
  5. For load, use any HTTP load tool you trust on a page of your own site, and send the site's host name in the Host header. Run it long enough to see steady numbers, and say what the tool and the machine were. Read memory from Task Manager or ps.

Run it on your own hardware and on the server you plan to use. If you get very different numbers, we want to hear about it.

Back to Compare, the changelog, or install it.