A cross-platform performance comparison of Finch v1.4.3 on two devices reveals a critical memory management issue on macOS:
Devices tested:
- MacBook Pro 2017 — i7-7920HQ, 16GB RAM, macOS
- Lenovo G470 (2011) — i5-2450M, 4GB RAM, Windows 10 Pro 22H2
Key Findings
| Metric |
MacBook Pro 2017 |
Lenovo G470 (15yr old) |
| Finch Renderer Memory |
5.7 GB (34.8% of 16GB) |
1.3 GB (32.2% of 4GB) |
| Throughput (avg) |
0.24ms ✅ |
1.56ms |
| 10K lines render |
6.6ms ✅ |
30.6ms |
| 20 concurrent tasks |
4.0ms ✅ |
15.6ms |
| User Experience Score |
2.8 / 5 |
3.8 / 5 |
Despite having a much faster CPU, the MacBook Pro delivers a worse user experience (2.8 vs 3.8). The root cause is the Finch Renderer process consuming 5.7GB of RAM — nearly 4.5x more than on Windows.
UX Breakdown
| Test |
MacBook Score |
G470 Score |
Notes |
| Quick chat |
2-3/5 |
3/5 |
Mac has input lag and scroll stutter |
| Large content |
2/5 |
3/5 |
Mac freezes during generation |
| App switching |
5/5 |
5/5 |
Both fine |
| Any task execution |
1/5 |
3/5 |
Mac UI freezes during tasks; G470 stays usable |
| File operations |
4/5 |
5/5 |
G470 near-instant |
| Scroll history |
3/5 |
4/5 |
G470 smoother |
Analysis
The MacBook's Finch Renderer consuming 5.7GB leaves only ~10GB for the OS and other apps, causing swap pressure and UI freezes during any task execution. On the G470, the Renderer uses only 1.3GB, leaving 2.7GB — proportionally similar (~33% of total RAM on both), but the absolute 5.7GB figure is excessive on a machine with only 16GB total.
The irony: a 15-year-old budget laptop runs Finch more smoothly than a 2017 flagship MacBook Pro. This suggests Finch's Electron memory management needs optimization on high-end machines, not just low-end ones.
Suggested Fix Direction
- Investigate why the macOS Renderer process allocates 4.5x more memory than on Windows for the same workload.
- Implement memory limits or garbage collection in the Renderer process.
- Consider whether GPU/WebGL memory allocations on macOS are leaking or over-allocating.
- Profile memory usage during long sessions — the 5.7GB figure suggests accumulated allocations that are never freed.
Full Report
Full test methodology and results: https://github.com/BlessedChild/finchstory/blob/main/finch-perf-test-comparison.md
Environment
- Device A: MacBook Pro 2017 · i7-7920HQ · 16GB RAM · macOS
- Device B: Lenovo G470 (2011) · i5-2450M · 4GB RAM · Windows 10 Pro 22H2
- App: Finch
1.4.3 (build 1070) — Desktop
Environment
1.4.3 (build 1070)— Desktop