High GPU usage (70–90%) on macOS with Codex app

Thanks for the bug report. I'm not aware of any other reports of this behavior, so we'll need to figure out what's different about your computer, configuration, or usage.

Is there anything special or noteworthy about the coding tasks or tools that the agent is running on your system? If you can repro this on demand, could you use the /feedback slash command to upload your logs and session details and then post your thread ID here?

Are you seeing especially large diffs during your session? We've had some reports of that resulting in poor performance, and we're working on some fixes.

Hello, currently experiencing the same issue as tavaresgmg. Submitted a bug report through /feedback

Yes, I am also experiencing significant slowness in the agent's operations in addition to the high GPU consumption. Even simple coding tasks feel sluggish when using the Codex app.

@etraut-openai Here is the thread ID from the feedback submission: 019c20f7-1823-79a3-a6d6-79281b34d349.

Experiencing the same behavior on an M4 Pro (24GB RAM). After viewing the diffs section once, I began seeing random GPU spikes between 60% and 100% while the agent is active. I suspect the diff view might be failing to release GPU resources even after being closed, as this usage pattern only started after I opened that section for the first time.

thread id: 019c22c9-3808-7fb3-a094-7aaf1985480e

Same here.

Thread ID: 019c2520-c5eb-72b3-93c3-8f900d73b962

Hardware: MacBook Pro M3 Max (14-core), 36 GB RAM

Observed issues:

• CPU usage consistently >120% even when idle in conversation
• Switching between threads takes 5+ seconds
• General UI responsiveness is sluggish (noticeable lag on clicks, scrolling)
• App becomes increasingly slower the longer it runs

Environment:

• macOS Sequoia 15.x
• Codex.app (latest version)

This is on a high-end M3 Max machine, so the performance should not be a hardware limitation. Happy to provide additional diagnostics or logs if helpful.

Same, my Mac dies of heat. 019c27bf-7b35-78e3-9c6d-755ae4583b24

I have the same problem, happens when chat is active. Might be animations causing excessive UI updates?

bb, same

It is eating through CPU and kicking in the fan so aggressively please fix this

Macbook Pro M5

When the codex is in the front the GPU util surge suddenly making my mac potato hot. When minimized to dock or covered by other window the GPU util drop to normal range.

<img width="320" height="272" alt="Image" src="https://github.com/user-attachments/assets/cf1eb8e0-5e3a-4ebd-a3af-df58064824d5" />

When the codex is in the front the GPU util surge suddenly making my mac potato hot. When minimized to dock or covered by other window the GPU util drop to normal range. <img alt="Image" width="320" height="272" src="https://private-user-images.githubusercontent.com/58298819/545348176-cf1eb8e0-5e3a-4ebd-a3af-df58064824d5.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.BLIlJE53dHqqZPRjNbwF-Ai852cTj51ZtnPufzYTMG0">

same behaviour

Same issue here, macbook air m4, even with small tasks and also when it's finished, it keeps the high usage.

I’ve given up , cancelled my subscription. Completely unusable app for me with the cpu melting inside my machine

It looks like you’re using an outdated Electron. They fixed this bug in September: https://github.com/electron/electron/issues/48311

@combdn, thanks for the suggestion, but we're using a recent version of Electron.

We're struggling to repro the high GPU utilization.

For those of you who are seeing this issue, does the problem go away if you switch to an opaque window (Settings -> Appearance -> Use opaque window background)?

@etraut-openai it doesn't. Only minimizing the app will reducing the gpu usage until you maximize it again.

@etraut-openai BTW codex monitor with codex app not running also have the same issue/behaviour, so maybe it is not codex app specific UI problem.

@ahmed-abouelkhaire, that's an interesting observation. Can you confirm that you see high GPU utilization when running CodexMonitor? Or is it CPU utilization? IIUC, CodexMonitor builds upon the Codex CLI, so CodexMonitor and Codex App have that in common, but I wouldn't expect the Codex CLI to be doing anything that exercises the GPU.

<img width="410" height="136" alt="Image" src="https://github.com/user-attachments/assets/15983b5d-cf87-451f-9d75-5eb7b0ac9747" />

The Codex app performance has been horrifyingly bad. It uses more CPU than my browser, there is noticeable input lag especially when typing and it makes the entire OS lag because it's so heavy. I have an M3 24GB macbook.

@ahmed-abouelkhaire, that's an interesting observation. Can you confirm that you see high GPU utilization when running CodexMonitor? Or is it CPU utilization? IIUC, CodexMonitor builds upon the Codex CLI, so CodexMonitor and Codex App have that in common, but I wouldn't expect the Codex CLI to be doing anything that exercises the GPU.

Mainly GPU utilization fluctuates much more, but the unexpected behaviour is that it is coming mainly from the WindowServer process, and this behaviour only triggers when I open Codex app or Codex Monitor. I tried the CLI inside my terminal (Ghostty), and everything is fine there.

The CPU fluctuates also but within the acceptable range compared to system utilization and other apps.

System details
M3 Pro, 18GB Memory, 26.2 OS Version

<img width="1037" height="80" alt="Image" src="https://github.com/user-attachments/assets/f5c9902f-8d5a-4525-89c6-5e47311675d4" />

<img width="346" height="140" alt="Image" src="https://github.com/user-attachments/assets/990e62bc-e3d2-4a41-96d3-05e740a8a925" />

Something is not right with the Electron, probably because WindowsServer still hits the first place on CPU usage.

Posting from https://github.com/openai/codex/issues/10885, I really hope this issue gets fixed soon.

I had Codex debug itself a bit from DevTools performance traces. If anyone else is having this issue, you can (partially) mitigate it by forcing reduced motion only for the Codex app by launching it like this:

/Applications/Codex.app/Contents/MacOS/Codex --force-prefers-reduced-motion

Codex also helpfully made an app shortcut so you don't need to launch it from the terminal:

cat > /tmp/codex-reduced-motion.applescript <<'APPLESCRIPT'
on run
  do shell script "osascript -e 'tell application \"Codex\" to quit' >/dev/null 2>&1; sleep 0.7; open -na \"Codex\" --args --force-prefers-reduced-motion"
end run
APPLESCRIPT

osacompile -o "/Applications/Codex Reduced Motion.app" /tmp/codex-reduced-motion.applescript

It helps because it at least it makes the text shimmer in "Thinking" go away (which alone consumes GPU like crazy), but unfortunately when the Codex app is streaming text, I still see the power usage jump from 5-6W to 10-14W.

Here are some notes by Codex, probably useless:
<img width="700" height="600" alt="Image" src="https://github.com/user-attachments/assets/d967b570-555a-4cf2-9694-795cde322883" />

Posting from #10885, I really hope this issue gets fixed soon. I had Codex debug itself a bit from DevTools performance traces. If anyone else is having this issue, you can (partially) mitigate it by forcing reduced motion only for the Codex app by launching it like this: `` /Applications/Codex.app/Contents/MacOS/Codex --force-prefers-reduced-motion `` Codex also helpfully made an app shortcut so you don't need to launch it from the terminal: cat > /tmp/codex-reduced-motion.applescript <<'APPLESCRIPT' on run do shell script "osascript -e 'tell application \"Codex\" to quit' >/dev/null 2>&1; sleep 0.7; open -na \"Codex\" --args --force-prefers-reduced-motion" end run APPLESCRIPT osacompile -o "/Applications/Codex Reduced Motion.app" /tmp/codex-reduced-motion.applescript It helps because it at least it makes he text shimmer in "Thinking" go away (which alone consumes GPU like crazy), but unfortunately when the Codex app is streaming text, I still see the power usage jump from 5-6W to 10-14W. Here are some notes by Codex, probably useless: <img alt="Image" width="700" height="600" src="https://private-user-images.githubusercontent.com/175479716/546193583-d967b570-555a-4cf2-9694-795cde322883.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NzAzODUwMTQsIm5iZiI6MTc3MDM4NDcxNCwicGF0aCI6Ii8xNzU0Nzk3MTYvNTQ2MTkzNTgzLWQ5NjdiNTcwLTU1NWEtNGNmMi05Njk0LTc5NWNkZTMyMjg4My5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwMjA2JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDIwNlQxMzMxNTRaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1iNDE4Zjk0MTlhY2FmZjA4MDc3MmU1YzE3YjdlYWIwNTc3NjdlOTJkZDRhYzVlYWNlY2YyMmJlN2U4ODJiYWU0JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.d6xlcPMINnRcaxE8rINcjUEHRTBDlC2SfDDKQuSDcT4">

yes, it helped! disabled motion as you recommended and now codex itself is almost not using any CPU, but WindowServer is still consumes a lot but at least my fan is not working

Any updates or ETA on this?

Also having this on a 24g m5 macbook pro. Here's the thread ID. My GPU (not cpu) sits at 65-75% with Codex open, even on empty tabs. Disabled Motion did nothing for me. I do not have this issue with the VScode extension, only the app.

I am on the latest OS, and have restarted / ran with only the Codex app open, same issue. Still happens in the backround, and with the transparent window off. I just updated to the latest version of Codex, but the issue remains.

019c353c-b1a7-7e73-b983-feabb0a1643a

Also having this problem.

Same story on windows with lates codex desktop.

I even turned on descret graphics, but it wont help.

<img width="1786" height="658" alt="Image" src="https://github.com/user-attachments/assets/d1a7d1f6-5d64-4204-a808-093d8d8b678f" />

<img width="1245" height="916" alt="Image" src="https://github.com/user-attachments/assets/642ce88c-2e58-4d14-8bd8-b2430384f9a2" />

Mac OS user's only on Tahoe experience this

I have the same problem on MacOs

I tested it and after adjusting the window width to the minimum, the GPU usage was reduced by half

MacBook Pro M1 Pro
Tahoe 26.2
Codex Version 260206.1448 (565)

Have also been experiencing latest version of Codex suddenly using all my CPU when trying to run one chat.

Only thing helping so far is:

• Click off the "IDE context" in each chat across any open projects, which keeps being on by default on any existing thread I have already or when starting a new thread.

Trying to go to 'Settings' then clicking 'Back to app' always causes it to spike CPU and freeze for a period even with this IDE Context off.

Mac OS user's only on Tahoe experience this

I'm on Sequoia and have this issue. Using codex desktop Version 260206.1448 (565)

> Mac OS user's only on Tahoe experience this I'm on Sequoia and have this issue. Using codex desktop Version 260206.1448 (565)

I'm surprised because many of the users on Sequoia don't have this issue. Probably someone on Tahoe also doesn't have this issue, so it's not for all users. But for some of us, the app becomes actually unusable.

We don't yet have a solid repro of this problem. Most OpenAI employees are using Macs, and we haven't seen any reports of high GPU utilization internally. We're looking for clues about what's common to those of you who are experiencing this issue. I wonder if it's a certain generation of Mac hardware, which could map to a particular video driver that is struggling with something that the Codex app (and extension) are rendering. Let us know if you have any other theories or clues.

To complicate matters, it appears that there are a couple of different issues at play here — one that's related to high GPU utilization and another that's related to high CPU utilization but not GPU. If you're seeing either problem, please be specific about what you're seeing and under what circumstances it occurs (e.g. when you take certain actions, work within certain projects, etc.).

Eric, thanks for looking into this. In case it helps, I will try more details here.

The problem seems to manifest more when I'm working on complex problems where the conversation thread can get large. I work on multiple projects at a time. If I quit codex and only open simpler projects with shorter threads, the problem does not exhibit itself. However if I open the larger threads the problem comes back.

I'm also curious as to whether you guys have telemetry enabled (or if I can enable telemetry and I can share some session ID's or trace ID's so that you guys can take a closer look).

It is very odd that nobody at OpenAI is able to repro. The bug seems to happen essentially on-demand for me (opening/working on certain threads).

!Image

Given that other users report this bug from M1 through M5 and several OS generations I am inclined to believe that this is indeed a software bug.

One thing to note is that I did have similar laggy issues when using the Codex VSCode extension as well (again, with longer threads). Perhaps the bug is caused by some library that is shared between the two? I am curious if anyone else had this issue with the VSCode extension as well.

I will note that I have perhaps been spoiled by Codex's exceptional ability to manage long context (as compared to other coding agents like Claude code) which is why I perhaps don't start new threads as often as I should. Perhaps the problem is exacerbated by this.

@Suhaibinator , thanks for the information. You mentioned "laggy behavior", but you didn't say whether you're seeing high GPU utilization. Can you clarify? As I mentioned, there are likely a couple of different issues here, so it's important that we tease them apart.

We don't yet have a solid repro of this problem.

@etraut-openai I think I found the root cause of the issue on my machine. Codex was showing a massive diff tree (+32,000) even after I committed my changes. Seems its including untracked files in the diff tree which includes files and folders autogenerated by IDEs and even codex itself as temp working spaces. When I added these files to my .gitignore and the diff tree went down to +3,000 performance and utilization went back to normal

@thegeronimo, I can understand why that would cause high CPU utilization. Can you clarify whether you are seeing high GPU utilization?

@thegeronimo, I can understand why that would cause high CPU utilization. Can you clarify whether you are seeing high GPU utilization?

I don't recall there being high GPU utilization as something that stood out but I also didn't focus in on that, I was mostly looking at CPU and memory usage. I can say the fans did spool up which is a rare event on my M2 macbook and eventually the O/S froze (unable to move windows around) at that points it would auto restart shortly after

Thing is, it’s a problem regardless of context length or task size.

Start a singular task, see CPU skyrocket. Minimise app = CPU drops again.

I have multiple macs of different generations of M chips and OS versions; all suffering this issue exclusively with codex, never had this before. Changing motion settings, transparency settings,… does nothing. Only minimising solves it, and the moment you bring it back to foreground the CPU spikes again, and fans go crazy.

Thing is, it’s a problem regardless of context length or task size.

@LasseRafn It might be the repo your working in itself, in the upper right hand corner of codex what does it show as the diff size?

@thegeronimo it sounds logical but thats not the case. a completely blank folder does the exact same thing.

low effort, newest model, create a simple file and it does ~70% CPU usage and temperature started rising.

minimize = CPU drops down to ~1%

https://github.com/user-attachments/assets/e1a25873-8e3e-45a3-bfe9-edbe1d230695

<img width="1049" height="277" alt="Image" src="https://github.com/user-attachments/assets/b612432a-8ddf-409c-8e53-f97812db1e08" />

<img width="959" height="125" alt="Image" src="https://github.com/user-attachments/assets/075937d4-b218-49fa-b622-b3e366765573" />

Some more data. What is odd is that the CPU utilization graph doesn't show elevated usage, but the entries in the table do. I have also shared my GPU utilization graph. It seems to be both?

@thegeronimo it sounds logical but thats not the case. a completely blank folder does the exact same thing.

@LasseRafn creating a new project might actually be creating a massive diff tree

@thegeronimo but there were no files and it already spiked, CPU goes crazy before any files are touched, in any project even a completely empty folder.

git wasn't even enabled in the folder in the video above, no diff to render:

<img width="1928" height="2262" alt="Image" src="https://github.com/user-attachments/assets/fcb8d5b8-5ebf-4651-85d7-c1c304547030" />

So I don't think its just the diff renderer, but its probably also having issues haha

<img alt="Image" width="1049" height="277" src="https://private-user-images.githubusercontent.com/42899065/546639007-b612432a-8ddf-409c-8e53-f97812db1e08.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.uGh6xXNpaOvGa83L03d9ICz9H8ZKvfmcy-F_DwFdd50"> <img alt="Image" width="959" height="125" src="https://private-user-images.githubusercontent.com/42899065/546639028-075937d4-b218-49fa-b622-b3e366765573.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.mXa3ktFOxvLQ3qO4dz0sG2hxAielUOM_l8thnLV2Hm0"> Some more data. What is odd is that the CPU utilization graph doesn't show elevated usage, but the entries in the table do. I have also shared my GPU utilization graph. It seems to be both?

this is actually strange, I have same problem too. Graph shows no usage but main list does

Experiencing extremely high GPU usage as well (CPU is mostly OK). Tried different things in the app, not sure if it helps, those are rough numbers partially from memory:

1. Looking at thread 019c31e8-b4a6-7810-a475-dda25928ef53 while this thread is doing work - ~40W GPU
2. Looking at another thread while this heavy thread is running - 10~20W GPU
3. Looking at fullscreen settings while this heavy thread is running - 0~10W GPU
4. Looking at the heavy thread once work has finished - 0~3W GPU

Found a way to reproduce most parts of it.

A little more context, when I am having this issue normally, even if Codex is in the background it occurs. I found a way to solve that issue, and vastly reduce usage while in focus.

I found two seperate ways to reproduce two different issues. These were both tested on an M5 Macbook pro, and an M2 Macbook pro, and I had identical results between both.

1. Codex using too much background GPU
2. Codex using too much GPU while focused

For issue 1, it can be fully resolved by making the window smaller. This took me from 65% GPU usage, to less than 1% while Codex was not in focus. My window size before was the whole screen (But not full screened), moving it to about 1/4 of the screen and placing it in the middle solved it. This took be from again 65%, to 1%

For issue 2, all testing was done with the smaller window, about 1/4 sitting in the middle of my screen, same as before. I noticed the following issues. Hovering over any chats spikes GPU usage to 30%. Hovering over the recommended cards (Build a classic Snake game in this repo) spikes me all the way to 75-85%

If I do not hover over anything, I sit around 3-15%, though I imagine this is still slightly too high at times.

Note: If you click into a repo, then start a new thread, a little animation plays on the codex logo above the cards. While this is playing, I sit at 65% nomatter what.

if you minimize the window GPU usage drops completely, even while working in the background. (mbp m5)

Tahoe 26.2 (25C56)M2 Max Mac Studio, external display - 6K display resolution scaled to a 4K Screen.

If I hide Codex, I'm at 10% GPU, 12W power consumption
If I fill the screen with the window, 50% GPU, 45W power consumption

• Same issue if I restart the app and don't open the diff pane or do anything.
• ("Use opaque window background" is on, fwiw)

The funny part is that if you connect your MacBook to an external monitor while using codex, GPU usage will drop completely!

@sinadarvi lmao I gotta test this tomorrow. (but sure, its a hardware problem, totally not a Codex problem /s)

The funny part is that if you connect your MacBook to an external monitor while using codex, GPU usage will drop completely!

This is an interesting observation. I have a math studio and therefore only have an external monitor. But I do get this problem! Perhaps it's not an external monitor issue, but a primary/secondary display issue?

Edit: Or... perhaps the issue is less apparent on non HiDPI external monitors?

for anyone else on this thread, I highly recommend using the OpenCode with Codex 5.3. I’ve been using OpenCode for a long time, and honestly, the official Codex app feels like a bit of false hype in comparison. You can connect your existing OpenAI subscription there, and in my experience, it harnesses the model much better. It seems to be more token efficient and provides far better suggestions/edits, likely due to its superior LSP integration, frequent type-checking, and background linting.

@kaanmertkoc if I wanted, I could just use Codex TUI which is very well optimized. The problem is that OpenAI is giving 2x usage if you use the Codex app specifically. (Besides, your post sounds a bit like an ad 😉)

Anyway, so far there's still no sign of OpenAI fixing this, unfortunately. Reduced motion helps a bit, at least, from https://github.com/openai/codex/issues/10432#issuecomment-3860314761

@hourianto, to clarify, the 2x usage rate is not limited to app usage.

so far there's still no sign of OpenAI fixing this

This issue is a high priority for us. Clearly, many users are hitting it, but we've struggled to repro the issue. There have been some good clues in this thread over the past day (thanks to everyone who has been posting!), so I'm hopeful that will lead us to a repro. We'll continue to investigate.

@etraut-openai thanks for clarification, I was very confused because "Tip: New Try the Codex App with 2x rate limits until April 2nd." sounded to me like it's only for the Codex app. And I do indeed hope you manage to quickly solve it, maybe you can get someone from this thread to do a remote debugging session or something like that?

@kaanmertkoc if I wanted, I could just use Codex TUI which is very well optimized. The problem is that OpenAI is giving 2x usage if you use the Codex app specifically. (Besides, your post sounds a bit like an ad 😉) Anyway, so far there's still no sign of OpenAI fixing this, unfortunately. Reduced motion helps a bit, at least, from https://github.com/openai/codex/issues/10432#issuecomment-3860314761

definitely not an ad! i just love the product and team! i even have one small contribution so might be little biased 😄

also like mentioned the usage is not limited to codex so you can use the 2x limits on opencode as well!

@etraut-openai probably a dumb question but are you guys using the codex app without external monitor?

a real battery killer

<img width="2058" height="1712" alt="Image" src="https://github.com/user-attachments/assets/4652fd85-e137-48e9-a3f2-12e683dba4a9" />

@etraut-openai Some more info that I hope will help with repro:

• In fullscreen mode I don't see elevated GPU usage, even at 6K. Perhaps a compositing rather than draw issue?
• In windowed mode I also don't see elevated GPU usage, if Codex is the only app.
• If I put a bunch of stock apps behind Codex (Music Safari, Finder, Mail and Photos), no elevated GPU usage
• If I put a Dia Browser Window behind Codex, I immediately get elevated GPU usage.
• If I then hide Codex _or_ Dia, GPU usage goes back down.

I don't know how get MacOS to report GPU usage on a per process basis, so I don't actually know if this is really a Codex issue or a Dia issue.

Other people with this issue: are you also Dia users?

@jonzhan I don’t know what Dia is so I’d say no for me

@jonzhan, thanks for those clues! If you disable translucency in the Codex settings, does the GPU usage drop? I'm wondering whether it's a compositing issue in macOS.

<img width="706" height="59" alt="Image" src="https://github.com/user-attachments/assets/c57341a3-76d5-4f4f-9e80-a33844766b33" />

@jonzhan, thanks for those clues! If you disable translucency in the Codex settings, does the GPU usage drop? I'm wondering whether it's a compositing issue in macOS.

At the time of testing, Use opaque window background was on (edit: wrote the opposite intially). I just toggled it a few times, and don't see any difference either way. 🤔

@jonzhan I don’t know what Dia is so I’d say no for me

@LasseRafn Can you also alleviate the symptom by hiding everything but Codex, in your case?

Out of curiosity to see if it's a Dia+Electron thing and not a Codex thing, I also launched every other Electron app I could find and put them behind Dia, with Codex hidden. No GPU spike in this case, though I have no RAM left heheh.

@jonzhan unfortunately for me the only thing that works is minimizing Codex. (I mean I didn't try hiding the activity monitor because thats what I use to check, so MAYBE..? I cant know for sure).

I try resizing, toggling panes, toggling settings, hiding other windows (except Activity monitor), the only times it stopped consuming a lot of CPU was when:

• Codex wasn't doing anything (finished its task)
• Codex was minimised

Can't really pinpoint what might be the cause though :(

Guys I am not sure but what I see - Codex don't eat much GPU, but CPU. Also the huge using amount is not showing here

<img width="406" height="106" alt="Image" src="https://github.com/user-attachments/assets/803a1343-5db1-4ae0-bb9d-798b113346f9" />

<img width="344" height="130" alt="Image" src="https://github.com/user-attachments/assets/95be3f67-64d2-498f-bbfd-9a19e06ba6ea" /> while actually hitting big usage with one simple thread

Suggest opening a separate ticket if you have CPU issues with Codex - best to focus this thread on the GPU usage originally reported!

I tried but they closed as duplicate of this one 😄 https://github.com/openai/codex/issues/10530

I'm experiencing the same issue. When the Codex desktop runs tasks, my GPU usage jumps to 90-100%, causing my mac to overheat. Let me know if there are any solutions or workarounds you've found

Machine type:

MacBook Pro M3 Max (48GB RAM)

@luyi404 I suggest minimising the window, that seems to work for me on all devices.

1. Start Codex task
2. Make sure notifications are on (so you know when to check back)
3. Minimise window
4. Come back when receive notification

Thanks! @LasseRafn This solution works for me too. It lowers my GPU usage to around 50%, which is still quite high but definitely better than 90%!

Machine type: > MacBook Pro M3 Max (48GB RAM)

Absolute beast of a machine - surely there's not many excuses for codex to make it hotter than literally anything else

I tried getting Codex CLI to help me debug what was happening live on the Codex app whilst clicking around and experiencing this freezing effect and fan kicking in issue.

---
I debugged it and captured concrete evidence from your machine.

Most important findings:

• Renderer memory footprint during sample: 3.9G (peak 4.4G) on a 16 GB Mac.
• Spike stack is dominated by Electron/V8 internals, with repeated v8::Isolate::LowMemoryNotification() signals.
• Codex logs show very frequent repeating loops near the spike window:
• repeated Skills/list request ...
• repeated git-origin resolution failures
• repeated desktop-notifications service starting
• Session log counts were high (120+ git-origin warnings, 100+ skills-list requests, 17 ResizeObserver loop errors).

Likely cause (inference): renderer-side performance regression under memory pressure, amplified by repeated background refresh loops (skills/git/notifications) when switching UI context.

!Image

Experiencing the same behavior on a mac air m2

<img width="538" height="115" alt="Image" src="https://github.com/user-attachments/assets/ae9c0f3a-6185-45c6-83fc-c3ea5e3b66eb" />

+1 my battery drains too quickly

m2 ultra here. Codex is so slow and unresponsive as the mac application. It' s unbearable.

api speed is awesome the app is very poor.

scrolling is very hard in chats with extreme delay and typing creates a big delay in seeing what you have typed. obviously i ded the obvious. restarted pc and the whole shebang.

<img width="325" height="295" alt="Image" src="https://github.com/user-attachments/assets/42597ffe-3f24-4757-b91b-02319fc3bf64" />

Noticeable improvement here. Thanks guys.

I think in the new update the CPU and fan issue has been fixed (Idk if entirely or not)

Oh. GPU usage is also also much better for me on Version 260210.1703 (602)

@jonzhan, thanks for the confirmation. Yes, we think the GPU utilization issue is fixed — or at least significantly improved — with the new release.

Thanks to everyone in this thread who provided feedback, clues, and repro steps.

@etraut-openai Nice! Can you perchance share in broad terms what was going on, for the curious?

GPU issue fixed for me, thank you!

The app is so unmanagably slow on mac, even on a fresh start the app is very laggy. it's close to impossible to give it new tasks. The AI responds quick the UI is poor. and when you press enter sometimes it does not appear in the app. Im on on M2 Ultra 64gb.

I had now 2 nights of no progress while ik took me hours to perpare this

@Karimrekiki which app version?

I had those symptoms previously but Version 260210.1703 (602) fixed them for me

Thank-you @etraut-openai for pushing the update. I just upgraded to Version 260210.1703 (602) and it has resolved the CPU spike, UI freezing and fan kicking in issue for me.

OK

@Karimrekiki which app version? I had those symptoms previously but Version 260210.1703 (602) fixed them for me

Thank you for the help. I noticed that I had a whole bunch that was uncommitted inluding package.json. This seemed to be the reason the app got slow. perhaps and edgecase but useful to fix.

This is the problem with using Electron. Why didn't they use Tauri or another better alternative to the horrifically bloated and almost universally-hated Electron?

@etraut-openai Thanks Eric, issue resolved.

This is not resolved. On an M4 mac mini, GPU goes into the 90s if the Codex window is in the foreground and it is displaying light animations (spinners for ongoing tasks, etc.).
Minimizing the window drops GPU usage.

@schniper, we've resolved many of the causes of high CPU usage in the app. If you're still experiencing high CPU usage, please open a new issue and provide repro steps.