High GPU usage in Codex Desktop on Windows while agent is running
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 45/100
- Issue type
- Bug
- Clarity
- Mostly clear
- Activity status
- Quiet
- Domain
- desktop-dev, operating-systems, performance
Research direction
Start by reproducing the issue in Codex Desktop on Windows and monitor ChatGPT.exe, dwm.exe, msedgewebview2.exe, and the child processes in Windows Task Manager during streaming and command execution. Compare GPU usage while the request runs and after it finishes; done means ordinary text and coding tasks no longer sustain unusually high UI GPU usage and return near idle after updates stop.
Written by the indexing model from the issue text.
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.803.41515
What subscription do you have?
Pro
What platform is your computer?
Microsoft Windows NT 10.0.26100.0 x64
What issue are you seeing?
Description
Codex Desktop on Windows causes unexpectedly high GPU usage while an agent is executing a request.
While Codex is working on a normal coding task, ChatGPT.exe uses around 40% GPU on my NVIDIA RTX 5070 Ti Laptop GPU.
The workload itself does not appear to be using the GPU. Processes launched by Codex such as node.exe, python.exe, esbuild.exe, and the Codex command runner all show 0% GPU usage.
The GPU usage appears to come primarily from the ChatGPT/Codex UI itself.
Environment
- OS: Windows 11
- GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU, 12 GB VRAM
- Integrated GPU: AMD Radeon integrated graphics
- RAM: 32 GB
- Codex: Windows Desktop App
- Model used when observed: GPT-5.6 Terra
- GPU driver: NVIDIA 32.0.16.1062
- Driver date: 2026-06-11
Steps to reproduce
- Open Codex Desktop on Windows.
- Open a coding project.
- Ask Codex to perform a normal coding task.
- Let Codex execute commands such as:
npm listnpm run buildgit diff
- Open Windows Task Manager.
- Go to Details or Performance → GPU.
- Observe GPU utilization while Codex is actively generating output and executing commands.
Actual behavior
While the Codex agent is working:
ChatGPT.exe: ~40% GPUdwm.exe: ~11% GPUmsedgewebview2.exe: ~5%node.exe: 0%python.exe: 0%esbuild.exe: 0%- Codex command runner: 0%
The NVIDIA GPU showed approximately:
- Overall utilization: ~40%
- Dedicated VRAM usage: only ~0.6 / 12 GB
- Video Encode: 0%
- Video Decode: 0%
After the Codex task finished:
ChatGPT.exedropped from ~40% to ~14% GPUdwm.exewas around ~25%node.exe,python.exe, andesbuild.exeremained at 0% GPU
This suggests that the GPU load is associated with the Codex/ChatGPT desktop UI or window rendering rather than the code being executed.
Expected behavior
For ordinary text/code generation and command execution, the Codex Desktop UI should not require 30–40% of a high-end discrete GPU for an extended period.
Some GPU usage for UI rendering/hardware acceleration is expected, but this level seems unusually high for displaying text, code blocks, command status, and diffs.
GPU usage should ideally return close to idle after UI updates stop.
Additional observations
The behavior is especially noticeable while Codex is:
- Streaming generated text
- Updating command execution status
- Showing terminal output
- Updating syntax-highlighted code
- Rendering diffs
- Automatically scrolling/updating the conversation
The AI model is not running locally, and the child processes performing the actual build/code operations show no GPU utilization.
This makes me suspect a possible excessive rendering/compositor loop or inefficient GPU-accelerated UI update path in the Windows Codex Desktop app.
Screenshots
I can provide screenshots showing:
- ~40% GPU utilization while Codex is executing a request.
ChatGPT.exeaccounting for ~40% GPU while Node/Python/build processes show 0%.- GPU usage decreasing after the request completes.
What steps can reproduce the bug?
Simply sending a command in Codex triggers this issue.
What is the expected behavior?
No response
Additional information
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.5k
- Avg merge
- 1m
- Merged PRs (30d)
- 1k
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from openai/codex
-
enhancement remote
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
-
bug CLI windows-os
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
-
macOS sandbox blocks hw.optional.arm64 sysctl, causing Flutter to misdetect Apple Silicon as x64 Openbug CLI sandbox
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
-
bug CLI TUI
Difficulty 2/5 1-3 hours Newbie friendliness 90/100
-
CLI config enhancement skills
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
kwakseongjae/auto-hwp#319 ·
-
area:cli bug filter-quality good first issue priority:medium
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
-
Difficulty 1/5 Under an hour Newbie friendliness 72/100
bevyengine/bevy#25861 ·
-
comp-datalake
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
ClickHouse/ClickHouse#121222 ·
-
A-linter
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
oxc-project/oxc#26863 ·