Codex on Windows 11 causes severe temperature spikes and system freeze

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #30021
• #30050

Powered by Codex Action

I've been having similar lag issues for multiple weeks now. It doesn't seem to matter whether I have "Windows Native" or WSL as the configured backend. Every time I use the Codex desktop app, it seems like it lags horribly and not only lags itself but also locks up my entire computer for a split second or so every single time it does something. Starting it along is extremely heavy and I'm not sure why and have tried archiving all of my chats thinking maybe it was just struggling to load a long chat history or something. My PC is a workstation able to play triple-A games at max graphics settings, so I don't think it's a problem with my computer itself and this seems to purely be something with the Codex desktop app as no other applications do this. I've also noticed a lot of graphical glitches with the Codex app that I'm again not seeing anywhere else.

Dunno if this is something weird with the graphics stack, a driver issue that's only triggered by Codex's app, or what, but it makes my computer extremely frustating to use when Codex is open. Constant microstutters across the entire system and CPU/GPU isn't even spiking.

I've tried reinstalling drivers, clearing files, running sfc and dism, and I've tried having Codex itself look into potential causes which it ended up blaming a bundled node_repl.exe but I don't know if it's actually that since I assume using WSL as the backend would cause it to run node within WSL... but I could be wrong on that. Either way, the UI is becoming basically unusable for me when no other application on my system is having this issue.

Dunno if this is related to the same issue others are having in this thread but figured it'd be better to just post here as a comment rather than making a whole new issue for it since it seems like I am not alone when it comes to the desktop app's performance problems.

Specs:
Processor: AMD Ryzen 9 9950X3D 16-Core Processor (4.30 GHz)
Installed RAM: 96.0 GB (93.7 GB usable)
Graphics card: NVIDIA GeForce RTX 5080 (16 GB)
AMD Radeon(TM) Graphics (2 GB)
Storage: 4.43 TB of 10.01 TB used
System type: 64-bit operating system, x64-based processor

Dunno if anyone at OpenAI is actually monitoring these threads, but here's a trace ID if you care to look into it: 019eb7e8-00dd-7740-8088-bc2d9e3eab95

Closing — can no longer reproduce on the latest build.

After updating to Codex Desktop 26.623.70822 (Jun 29, 2026) on the same machine (Windows 11, Intel i9‑12900HK, Minisforum NAB9), I can no longer reproduce the pathological idle/thermal behavior that prompted this issue.

I re-tested with external monitoring from a separate Ubuntu collector reading this Windows host over SSH, so the figures are not produced by the app under test.

Across several samples:

• Codex Desktop idle/background now sits around ~2.6% of one core / ~840 MB RAM, with temperatures returning to normal idle/light-load range.
• In mixed real use with browser + Codex + another Electron app open, Codex load is comparable to other active Chromium/Electron apps.
• The temperature now oscillates with CPU/GPU activity and drops again when load falls; I no longer see the monotonic runaway / freeze behavior from the earlier build.
• I also compared with Claude Desktop side-by-side: renderer/GPU load appears in whichever app is actively rendering/streaming, not uniquely in Codex.

Revised conclusion: the older build may have had pathological behavior on this machine, but the latest build is substantially improved. I would no longer say that Codex Desktop overheats the machine just by being open.

Closing as no longer reproducible on 26.623.70822. Happy to reopen with fresh external-monitor data if it returns.

<img width="1885" height="407" alt="Image" src="https://github.com/user-attachments/assets/cf12cfb9-1ccc-4476-9367-69c7735908fd" />

Reopening — the thermal burst behavior is back on a build newer than the one where it seemed fixed (fresh external-monitor data, as promised).

Same machine (Windows 11, Intel i9-12900HK, 32 GB, Minisforum NAB9), same methodology as my closing comment: an external Ubuntu collector polls this host over SSH, so the monitoring itself isn't skewing the numbers.

What I observed today (2026-07-02, ~19:01 CEST / 17:01 UTC):

• With Codex Desktop running, package temperature was flat at ~50 °C for ~8 minutes.
• At 19:01:04 CPU jumped to ~88% with the GPU 3D engine at ~55% simultaneously; package temperature ramped from ~52 °C to 89 °C within minutes and stayed in the 80–89 °C band for the following ~10 minutes (see graph).
• Per-process snapshot during the episode: two Codex.exe renderer processes had accumulated 72 s and 68 s of CPU time (~340–350 MB RAM each — memory is fine, this is a CPU/GPU issue).
• The pattern is now bursty rather than monotonic: sampling again a few minutes later, the same processes were back to ~0.4% CPU / 0% GPU, then temperature oscillated with further bursts while Codex stayed open. So it is improved vs. the runaway I originally reported, but the bursts are still severe enough to push the package to 89 °C.

Why this hits this machine hard: it only has the Intel Iris Xe iGPU (no dGPU). GPU load lands on the same die and power budget as the CPU, so whatever the renderer is doing on the 3D engine translates directly into package temperature.

Installed build: 26.623.81905, released Jul 1, 2026 (MSIX package OpenAI.Codex 26.623.11225.0, Electron 149.0.7827.197). This is newer than build 26.623.70822, where I could no longer reproduce the issue on Jun 30 — so this looks like a regression (or an intermittent issue that 70822 merely made less frequent).

Possibly related: #30021 (idle CPU core burn while the window is visible) and also #30721 (Codex ... appears to trigger Defender activity).

<img width="917" height="448" alt="Image" src="https://github.com/user-attachments/assets/d31cec83-a625-4f56-8a90-a054b70f3f14" />

Environment:

Upgraded to Version 26.623.101652 (Released Jul 3, 2026)

Issue Description:

It appears that simply keeping Codex open triggers the system's antimalware process. This causes sustained high CPU usage, which consequently raises the system temperature.

<img width="1915" height="1040" alt="Image" src="https://github.com/user-attachments/assets/68f7989c-bb69-49eb-a6dc-097f67778d35" />

<img width="531" height="137" alt="Image" src="https://github.com/user-attachments/assets/c7d0e6c1-23fd-48e4-9921-182ee3875253" />

Observation after closing:

As shown in the screenshot below, immediately after completely exiting Codex (including closing it from the system tray), the antimalware process consumes less CPU.
<img width="553" height="156" alt="Image" src="https://github.com/user-attachments/assets/8af180f2-124c-4559-8468-31f65c36b603" />

Consequently, overall CPU usage normalizes (from 15% to 2%) and the temperature drops by about 20°C.

<img width="1915" height="558" alt="Image" src="https://github.com/user-attachments/assets/7b4f7bd2-1561-4944-b892-ad648a82e468" />

Behavior at launch:

Observe the significant spike in system resources during the first few seconds of launching Codex: CPU jumps from 3% to 60%, GPU from 3% to 40%, and temperature from 49°C to 73°C.

<img width="947" height="653" alt="Image" src="https://github.com/user-attachments/assets/b0a38cbc-a4ad-45bd-84ca-d73d961ec6b7" />

Comparison with Claude

Out of curiosity, I compared this behavior with the Claude Desktop app. As shown below, when actively using Claude, the GPU usage remains consistently higher than CPU usage, and the overall temperature stays 10°C to 20°C lower.
<img width="1589" height="1009" alt="Image" src="https://github.com/user-attachments/assets/3ee467e0-443e-4a44-a8a9-5a5fc448ece8" />

Codex is now ChatGPT app.

ChatGPT
Powered by Codex & OWL
Version 26.707.31428

Released Jul 10, 2026
© OpenAI

<img width="944" height="870" alt="Image" src="https://github.com/user-attachments/assets/7e3ed1d3-8703-4871-a675-565dea2c54a3" />

There is even extra cooling via external conditioner!

Look below what happens when I quit ChatGPt app

<img width="942" height="872" alt="Image" src="https://github.com/user-attachments/assets/16a35d0b-b6a4-41fa-a5d0-62c588690e43" />

Retest with new data (2026-07-15) — reproduces at launch, idle, with numbers

Retested today on the same machine (Minisforum NAB9, i9-12900HK, 20 logical threads, Windows 11, 32 GB). Two relevant changes since the original report:

• The third-party AV (ESET) present on 25 June was uninstalled on 28 June — this retest ran with stock Windows Defender only, so the AV confounder is mostly gone.
• The case was open with air conditioning blowing on it during the retest — and it still spiked.

What happens at launch, doing nothing

Simply launching the app — no prompt, no task, idle on the "What should we work on" screen:

• App processes started at 18:15:11. Thirty seconds later, total CPU hit 44% and the hottest core (Core Max) jumped from a ~48–60 °C baseline to 78 °C, peaking at 87 °C during a first launch a few minutes earlier (open case + AC, mind you).
• Within ~4 minutes of startup, the ChatGPT + codex processes had accumulated ~80 CPU-seconds combined, front-loaded in the first minute — all unprompted.
• After settling, the main ChatGPT process kept intermittent background CPU (spot samples over 4–8 s windows ranged from ~12% of one thread down to near zero).

Why "low CPU in Task Manager" is misleading here

On a 20-thread CPU, a single thread boosting on a P-core (~4.9 GHz) reads as ~5% total CPU yet pushes the hottest core past 80 °C. Total CPU % hides this; the numbers above come from per-process CPU-time deltas and per-core counters, not Task Manager.

Baseline comparison on the same machine

Claude (desktop/CLI agent sessions) reaches comparable temperatures on this machine only under sustained agentic workload (measured today: ~85% of one thread across its processes while actively working). Codex/ChatGPT desktop reaches them by being launched and left idle. Same silicon, same chassis, same day — the differential is the app.

Update on the freeze

The complete system freeze from the original report has not reproduced since removing ESET, so that part was likely the AV interacting with the app's startup I/O. The thermal spike at launch, however, reproduces consistently without any third-party AV and is specific to this app.

Happy to provide the sampling commands or rerun with any diagnostic build.