Performance

How Otty stays fast, what to measure, and what to tune.

Render pipeline

• macOS: native Metal renderer, one GPU layer per pane.
• Windows (planned): native DirectX 12 renderer.
• Subpixel-aware glyph caching, so a repeated glyph is rendered once and reused.
• Inactive tabs can release their renderer resources entirely — see freeze-inactive-tab below.

Memory

• Sparse-row grid storage — only cells that have ever been written are allocated. Empty scrollback rows are free.
• Scrollback capped per pane:

scrollback-rows = 10000        # default

• Inline images use a small atlas with eviction; very large or rare images bypass.

Benchmarks

Throughput measured with vtebench — each test streams a fixed ~1 MiB of escape sequences and records how long the terminal takes to parse and render it. Lower is faster, and each row's fastest result is bold.

Take these numbers with a grain of salt

This is a sample benchmark to show Otty is fast enough — not a claim that Otty is definitively faster than Ghostty.

• It was run on my personal MacBook with other apps open in the background. This is not a controlled test environment.
• The absolute results don't mean much. The first debug build of Otty I ran this on was ~100× slower than Ghostty, yet I never felt any lag using Otty for real work.
• The benchmark is very different from real usage. The unicode test, for instance, hammers the terminal with Chinese characters I can't even read — you'll never see that density of CJK in one screen during actual work.
• A terminal can also look fast here because it has bugs or hasn't implemented a feature yet. Fewer things to do is not the same as doing them better.

Treat this as a rough sanity check, not a leaderboard.

Test environment

• Machine — MacBook Pro, Apple M5 Max (18-core), 64 GB RAM
• OS — macOS 26.5.1 (build 25F80)
• Terminals — Otty 1.0.2 · Ghostty 1.3.1 · Terminal.app 2.15 (470.2)
• Date — 2026-06-11
• Release builds, identical window size, run back-to-back in one sitting.

Warm (steady state)

Caches and glyph atlas warmed up — representative of day-to-day use.

Benchmark	Otty	Ghostty	Terminal.app
dense_cells	7.2 ms	23.6 ms	24.4 ms
medium_cells	22.8 ms	24.8 ms	34.9 ms
sync_medium_cells	23.0 ms	29.5 ms	38.9 ms
unicode	23.1 ms	23.3 ms	27.3 ms
scrolling	21.1 ms	39.3 ms	140.2 ms
scrolling_fullscreen	21.0 ms	48.3 ms	143.6 ms
scrolling_top_region	41.0 ms	39.6 ms	117.3 ms
scrolling_top_small_region	19.4 ms	39.2 ms	117.1 ms
scrolling_bottom_region	19.7 ms	39.1 ms	116.3 ms
scrolling_bottom_small_region	19.8 ms	39.1 ms	116.5 ms

At steady state Otty is fastest on 9 of 10 tests — about 3× faster than both rivals on dense_cells, and 2–6× faster on scrolling, where Terminal.app falls to ~120–140 ms while Otty holds ~20 ms. Otty and Ghostty are within a millisecond on unicode and scrolling_top_region (the one test Ghostty edges out).

Cold start (first run)

First pass right after launch, before any caches warm up.

Benchmark	Otty	Ghostty	Terminal.app
dense_cells	23.8 ms	62.3 ms	70.1 ms
medium_cells	32.8 ms	64.8 ms	67.9 ms
sync_medium_cells	32.4 ms	55.2 ms	60.9 ms
unicode	30.0 ms	57.7 ms	51.3 ms
scrolling	53.7 ms	130.5 ms	168.2 ms
scrolling_fullscreen	56.7 ms	147.7 ms	178.9 ms
scrolling_top_region	68.3 ms	136.3 ms	159.2 ms
scrolling_top_small_region	42.6 ms	127.3 ms	150.8 ms
scrolling_bottom_region	41.7 ms	125.4 ms	153.2 ms
scrolling_bottom_small_region	41.9 ms	126.4 ms	155.6 ms

Reproduce it

git clone https://github.com/alacritty/vtebench
cd vtebench && cargo build --release
# Run inside the terminal you want to measure, then compare:
./target/release/vtebench --bench-path ./benchmarks -o results.txt

Numbers vary by machine, GPU, and font — what matters is running all terminals on the same setup in one sitting.

Profiling and reports

otty config get memory-report

In-app, Window → Memory Report breaks down per-window / per-tab memory, atlas size, and grid allocations.

Crash and performance reports under ~/Library/Application Support/Otty/Crash/.

See also

• Configuration Reference
• Unicode and Text Styles