#Performance Benchmark

This page records the performance benchmark results for RenMori 1.1.0: the tests run inside a real Ren'Py 8.5.2 engine, build the same asset set as unprotected, baseline-protected and fully protected distribution builds, and time every asset read through the engine's own loading path, renpy.loader.load, so the figures reflect the shipped runtime rather than a standalone crypto library.

The benchmark script is tools/benchmark.py in the repository, and the raw results behind this page are in docs/benchmarks/2026-10-31-apple-m1.json.

#Summary

• With full protection, in-engine decryption throughput is 567 to 1529 MB/s depending on file size
• Reading a 2 MB CG takes about 0.95 ms longer with full protection than without; for reference, one frame at 60 FPS lasts 16.7 ms
• A random seek plus a 64 KB read in a video takes 0.086 ms (median) with full protection
• A lint run, which loads and parses every script, goes from 0.52 s to 0.67 s
• The game/ directory of a protected build grows by only 0.18%
• Protecting 319 MB of assets at build time takes about 13 s with full protection

#Test environment

#Method

#Workload

The Ren'Py SDK sample project the_question is extended with a synthetic asset set sized like a commercial visual novel, 812 files in all, bringing game/ to 319.3 MB including the sample's own files; the synthetic assets are random bytes, as incompressible as encoded images, audio and video.

#Measurements

MB on this page means 2²⁰ bytes.

#Results

#In-engine read throughput

In MB/s; higher is better.

#Per-file read time

In ms, the time to read one whole file.

Video is streamed by the engine rather than read as a whole file, so the latency figures in the next section apply to it.

#Video streaming latency

In ms, median.

#Script loading

the_question contains 52 compiled scripts, and the three builds differ by no more than 0.15 s.

#Build time and size overhead

Build time is spent only on the developer's machine or in CI and never affects players; full protection is performed by the closed plugin and includes extra work such as per-build structural polymorphism, so it takes longer than baseline protection.

#Chunk size

These figures come from the build tool's Python with the cryptography backend and serve to compare chunk_size trade-offs; they do not represent absolute in-engine performance.

Larger chunks raise sequential throughput and lower size overhead but make each random read decrypt more data; the default of 64 KB balances the two and rarely needs changing.

#Interpreting the results

Full protection uses the native runtime provided by the closed plugin and is the default for distribution builds; baseline protection uses the reference reader written in Python, with roughly 1/14 to 1/6 of the throughput of full protection, and suits only situations where the plugin cannot be used (see Protection Strength and the Closed Plugin).

The unprotected figures above approach memory-copy speed because the assets were already in the operating system's page cache during the test; this setup removes disk performance from the comparison so the results reflect only the CPU cost of decryption, and on a player's machine the first read is usually bound by disk speed, which narrows the relative gap between protected and unprotected builds.

#Limitations

• Frame time during real play is not measured; this page covers asset reading and decryption only, not image decoding, audio decoding or rendering
• The synthetic assets are random bytes that the engine reads but does not decode
• The tests run with a warm page cache and exclude cold-start disk I/O
• Only one Apple M1 device was tested; figures for Windows, Linux and Steam Deck have not been measured yet

#Reproducing

This needs the Ren'Py SDK and the RenMori source repository; full protection needs the closed plugin, which is downloaded automatically on the first run.

git clone https://github.com/MewBaka/RenMori.git
cd RenMori && pip install -e .
python tools/benchmark.py --sdk /path/to/renpy-8.5.2-sdk \
    --game /path/to/renpy-8.5.2-sdk/the_question --json result.json

Absolute figures vary between devices, so compare the relative results of the three builds on the same device.