Severe scroll stuttering in long Codex chats on Intel MacBook Pro

Summary

Scrolling a Codex conversation in the ChatGPT desktop app is severely stuttery even while the Mac has no memory pressure, swap usage, thermal throttling, or heavy background workload. The issue persists after fully restarting the app and also when connected to power.

Environment

• ChatGPT/Codex desktop app: 26.730.61639 (build 6234)
• macOS: Tahoe 26.4.1
• Hardware: MacBook Pro 16-inch (2019)
• CPU: 2.4 GHz 8-core Intel Core i9
• RAM: 64 GB DDR4
• Integrated GPU: Intel UHD Graphics 630, 1536 MB
• Discrete GPU: AMD Radeon Pro 5500M, 8 GB
• Built-in Retina display: 3072 × 1920
• Codex surface: local project chat

Steps to reproduce

1. Open a Codex project conversation containing multiple messages, code blocks, terminal output, and an image.
2. Scroll vertically through the conversation using the trackpad.
3. Observe strong frame drops and visible stuttering.
4. Quit the app completely, reopen it, connect the Mac to power, and repeat.

Actual behavior

The chat scrolls with severe stuttering. During scrolling, measurements from Activity Monitor/top showed approximately:

• Codex GPU service: peaks around 25% CPU
• WindowServer: approximately 17–21% CPU
• Codex renderer: approximately 270 MB RAM
• System swap: 0
• No macOS thermal or performance warning was recorded

The UI was being rendered through Intel UHD Graphics 630 despite the machine also having an AMD Radeon Pro 5500M.

Expected behavior

Conversation scrolling should remain smooth on this hardware, particularly with 64 GB RAM, no memory pressure, and no thermal throttling.

Additional notes

The problem is reproducible after an application restart. No private session transcript or application logs are attached because they may contain project/server information. I can provide sanitized diagnostics if maintainers specify what is needed.