Ep 8: Performance Benchmarks — Comparing 6 Terminals

Uniform hardware, uniform tests. A four-dimensional comparison across pure output throughput, Agent load handling, memory usage, and startup speed.

Methodology

Principle: Same machine, same shell, same test scripts. The only variable is the terminal itself.

flowchart TB
    subgraph Test Environment
        HW[MacBook Pro M3 Pro
36GB RAM
macOS 15.x]
        SH[zsh 5.9
Unified .zshrc]
    end

    subgraph Test Subjects
        T1[Ghostty]
        T2[cmux]
        T3[iTerm2]
        T4[Warp]
        T5[Terminal.app]
        T6[tmux inside Ghostty]
    end

    subgraph Test Scenarios
        S1[Scenario 1: Pure Output Throughput]
        S2[Scenario 2: Mixed Output Scrolling]
        S3[Scenario 3: Real-world Agent Load]
        S4[Scenario 4: Resource Usage]
    end

    HW --> T1 & T2 & T3 & T4 & T5 & T6
    SH --> T1 & T2 & T3 & T4 & T5 & T6
    T1 & T2 & T3 & T4 & T5 & T6 --> S1 & S2 & S3 & S4

Test Scenarios

Scenario 1: Pure Output Throughput — cat large file

# Generate a 10,000 line test file
seq 1 10000 > /tmp/test-output.txt

# Measurement: Time from starting cat until rendering is complete
time cat /tmp/test-output.txt

Scenario 2: Mixed Output + High-speed Scrolling — find full disk search

# Measurement: Fluidity during high-speed scrolling of many file paths
time find /usr -name "*.dylib" 2>/dev/null

Scenario 3: Real-world Agent Load — Claude Code executing a large Phase

# Start Claude Code and execute an operation that generates massive output
claude
# Input: /gsd-execute-phase 1
# Observation: Fluidity of output, lag presence, and scroll smoothness

Scenario 4: Multi-tasking Resource Usage

# Open 10 tabs/sessions, each running 'sleep 9999'
# Measurement: Total memory consumption

Test Results

Based on community benchmarks (VibeHackers, Mitchell Hashimoto benchmark, HN community):

Rendering Speed:

★★★ = Fluid/No lag | ★★ = Slight delay | ★ = Noticeable lag

Key Data Points:

• Ghostty vs. iTerm2: Approx. 3x speed difference (VibeHackers benchmark).
• Ghostty vs. Warp: Approx. 2.5x speed difference.
• Ghostty ≈ cmux: cmux uses the Ghostty rendering engine.

Resource Usage:

Attention Management (Efficiency metric for parallel Agents):

Conclusion

Pure Speed → Ghostty ≈ cmux > iTerm2 ≈ Warp > Terminal.app
Multi-Agent Mgmt → cmux > tmux > Warp > iTerm2 > Ghostty > Terminal.app
Remote Persistence → tmux > Others (The only one with detach/attach)
AI Integration → Warp > Others (The only one with built-in AI)
Resource Usage → tmux > Terminal.app > Ghostty > cmux > iTerm2 > Warp

There is no single "fastest" terminal; there is only the "most suitable" terminal for your specific scenario.