Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
macOS Tahoe + WindowServer / memory pressure over long uptimes — anyone else seeing this?
- Thread starter RumorConsumer
- Start date
- Sort by reaction score
Speaking as a software developer, a good 75% of OP's post is pure nonsense and entirely meaningless. ChatGPT did not help you out here, it diluted your message with meaningless fluff and confused you into false confidence about what's going on here.
For one thing, your memory behaviour is perfectly normal and not at all something that would cause problems. Unused memory is wasted memory so macOS will fill it up with things that may be useful later. It may even compress some of this cached stuff because it's much much faster to decompress it from memory than pull it in from disk, and it gives you more memory capacity to work with. If you have plentiful enough free disk space it may even choose to use your swap space to put actually running process memory in swap if it is determined to be less likely to be accessed than something else and more beneficial to keep the caches than use the RAM for relevant memory mappings.
Memory management is fairly sophisticated these days and it's all in favour of performance, and really the only metric the average user should bother with is the memory pressure.
Window Server's CPU usage may or may not be high depending on how you're reporting it. If it's using up to 50% of a single E-core then it doesn't really matter all that much. If it's 50% of the whole CPU die, that's a different story. CPU usage percentages are commonly reported in both full chip load and per-core loads.
If you have not changed anything from default I believe Activity Monitor, out of the box, reports single-core percentages on the individual items in its list. This means that a single, all core workload, on an M1 Max could be 10,000%.
I frequently have uptimes of several months on my own M1 Max machine (although it is rarely used) and several weeks on my M4 Pro and M4 Max machines. There are things you can do to cause leaks in WindowServer, but it is not a common problem and your data doesn't fit with that being a culprit for your experience.
Have you tested if you actually need to do a full reboot or restarting Safari is enough?
If your problem persists after restarting Safari only, try the following command in a Terminal
sudo killall Dock
It will ask for your password, but not show it as you type it, this is intended.
After executing the command the Dock and desktop background picture will briefly disappear as the Dock process and its children restart, this includes some of the compositor effects like Spaces.
For one thing, your memory behaviour is perfectly normal and not at all something that would cause problems. Unused memory is wasted memory so macOS will fill it up with things that may be useful later. It may even compress some of this cached stuff because it's much much faster to decompress it from memory than pull it in from disk, and it gives you more memory capacity to work with. If you have plentiful enough free disk space it may even choose to use your swap space to put actually running process memory in swap if it is determined to be less likely to be accessed than something else and more beneficial to keep the caches than use the RAM for relevant memory mappings.
Memory management is fairly sophisticated these days and it's all in favour of performance, and really the only metric the average user should bother with is the memory pressure.
Window Server's CPU usage may or may not be high depending on how you're reporting it. If it's using up to 50% of a single E-core then it doesn't really matter all that much. If it's 50% of the whole CPU die, that's a different story. CPU usage percentages are commonly reported in both full chip load and per-core loads.
If you have not changed anything from default I believe Activity Monitor, out of the box, reports single-core percentages on the individual items in its list. This means that a single, all core workload, on an M1 Max could be 10,000%.
I frequently have uptimes of several months on my own M1 Max machine (although it is rarely used) and several weeks on my M4 Pro and M4 Max machines. There are things you can do to cause leaks in WindowServer, but it is not a common problem and your data doesn't fit with that being a culprit for your experience.
Have you tested if you actually need to do a full reboot or restarting Safari is enough?
If your problem persists after restarting Safari only, try the following command in a Terminal
sudo killall Dock
It will ask for your password, but not show it as you type it, this is intended.
After executing the command the Dock and desktop background picture will briefly disappear as the Dock process and its children restart, this includes some of the compositor effects like Spaces.
You need to reinstall completely and manually move stuff over manually. Even if you have a Time Machine backup you can't restore from New OS > Old OS
Outside of your statement regarding ChatGPT, I totally disagree.
There have been fundamental changes to Tahoe vs Sequoia and I'm too lazy to exactly check, however this is where it lands through my testing over the last several months on Tahoe including the latest .3 Beta 3.
1. They are definitely changing something in the graphics subsystem. I see updated drivers especially for AMD. Intel and M series folks report glitches and bad performance overall for the GPU.
2. Memory managment including WindowServer changes are huge, there is a lot of inconsistency and bad performance across the board and it just "feels different" than Sequoia. Simple tasks take longer. Idle times eat up memory and CPU/GPU cycles for no apparent reason, even after turning off transparency for Liquid Glass.
We are already at .3 release and the performance is horrendous (especially on Intel).
I have a 2019 higher end Mac Pro and an M1 Max MacBook Pro and an M4 Pro MacBook Pro and all of them suffer one way or another from bad performance.
I've been using Macs since the 90s so I know exactly at what point macOS releases get better on each release, as well as remembering OS's and performance improvements ie Leopard vs Snow Leopard.
Yeah and @casperes1996 your point is taken about AI. I was trying to get it to help present the data and ultimately Im not worried about a spiking e-core I totally agree. It's much more this issue of what SDAVE is pointing to. The machine feels like it's doing significantly more work pretty much at every turn. I have another identical machine as a task machine and it's still on Sequoia and feels significantly faster. App launch times, general responsiveness, task switching all is improved. Tahoe feels like it assumes later hardware in a way that Sequoia did not. Something is different.
What's changed is there is a boatload of background processes now.
There have always been a large number of processes running, but now it's quite a bit more.
However, with my M1 MBP and mini - while my usage patterns don't match yours - I'm not having any responsiveness issues at all. None.
If you want to clear a lot of the inactive memory use, you can use sudo purge in the terminal. It may have an impact for you, or it may really do nothing of value. FYI I do see memory growth for WindowServer.
There have always been a large number of processes running, but now it's quite a bit more.
However, with my M1 MBP and mini - while my usage patterns don't match yours - I'm not having any responsiveness issues at all. None.
If you want to clear a lot of the inactive memory use, you can use sudo purge in the terminal. It may have an impact for you, or it may really do nothing of value. FYI I do see memory growth for WindowServer.
The way I read this your points are not in disagreement with what I said.
1)
I do not dispute there are changes to the graphics stack. For one thing Tahoe introduced Metal 4 which absolutely requires driver updates and is a very different API.
There may be issues and optimisation left to be done, sure. But overall my tested graphics performance on machines from M1 to M4 has actually improved (very slightly) on Tahoe under intense workloads
2)
I did acknowledge it's possible to expose a leak in WindowServer; In some CrossOver uses I've had it hit 20GB+. But it's non-trivial to trigger and doesn't occur in regular use. Furthermore I don't believe the issues OP is experiencing relates to memory at all. Not resident memory, not swap and not compressed memory. It's all a red-herring IMO.
Enabling "reduced transparency" also shouldn't have any effect at all, because the complete transparency and lighting shader code is still executed, and in fact it adds another post processing and alpha blending step *after* regular UI calculation to reduce the transparency effect. This has at least been how it's worked for a long time including at least 26.0 - For modern hardware it also really shouldn't break sweat at all to render a bit of transparency.
macOS 26 has bugs and issues, as all operating systems do. Some I particularly find annoying is that the Spotlight Apps view, when triggered with the trackpad gesture, can sometimes glitch out and when it does, it refuses to open at all and requires hitting escape followed by triggering Spotlight with cmd-space before it will work again. And also the OS can get into a state where swiping between spaces in Mission Control blacks out your wallpaper during the swipe. killall Dock will temporarily fix it.
I have not really experienced responsiveness or performance problems myself, in fact overall I think Tahoe runs ever so slightly snappier than Sequoia, but I am not disregarding the experience of those who have experienced such issues. I just feel very confident it is not related to memory behaviour, and the memory numbers we actually got in the original post are all well within reasonable ranges I think.
I do feel fairly confident that it's possible the issue can be resolved by removing some extension, relaunching a particular application or service rather than the whole system or similar.
I'm sure Apple would love to fix whatever root cause is behind all of this (assuming it's even their software at fault and not something third-party), but I can also say that the type of bug report that always gets buried at the bottom of my TODO at work is "after a long period of time x happens" or "thing randomly happens" with no further details. It can still be valuable details but more information than or concrete details is practically a necessity to get anywhere with the issue, especially if your own QA team or yourself can't replicate it.
And performance is also a bit tricky to discuss in bug reports. A lot of bug reports I get in are also just a mismatch of expectations. I'm not saying anyone in this thread does the following but I've gotten in tasks that just say "content is slow to load when on bad network".... Yes... It is... What on Earth am I expected to do about it? The content is already compressed and minified. A higher end 2019 Mac Pro or Mx Max with 32GB of RAM are machines that can deal with a lot but if Apple gets bug reports saying xyz is slow, I bet the reports need a lot of noise behind them before they get prioritised, because they probably get a decent chunk of those from folks running the OS off of a 5400RPM hard drive and doing a recursive file scan of everything on their 8TB drive or something like that.
And keep in mind that performance issues are not universal. Even with long uptimes and Safari constantly running with a variety of pages
I think a lot of Apple apologists are doing the same thing they have done for decades.
Tahoe just overwhelms the system for no apparent reason whatsoever. Liquid Glass/translucency or whatever effect they use should NOT impede performance. It is even taking away precious GPU/CPU cycles from M series chips, which would rather go to more important things like apps that are running in the foreground.
I also do significant work in Google Chrome (Google Suite and Web App based workflows), while folks may say Chrome is a GPU/CPU/Memory hog, I disagree, because my same workflow is perfectly fine on the same machine(s) in Sequoia. Even if you quit Chrome and only run pro software like Creative Cloud (Photoshop, After Effects, etc) there is a significant performance drop. My timeline playheads just jump around erratically and tasks like previews in AE are just slow which breaks my concentration/workflow and makes me really angry.
This is simple A/B testing. Are we saying that later versions of Tahoe won't make performance better? No, what we're saying is no amount of Liquid Crass is worth the extra horsepower that's being eaten away here.
I remember when Leopard came out I had a higher end G5, and even by .3 it didn't run this bad or slow down my workflow. I use my Macs professionally for a living and they need to do certain things with my workflow.
I'm so glad I went back to Sequoia. I wish I had a backup of it so I didn't just waste 2 days restoring it manually. Maybe it was a blessing in disguise though, I made a fresh install and move stuff over manually and I hadn't done a fresh install of macOS in years.
I guess it depends on what you consider "testing" under your workflow. I am not sure how you do your testing or what your workflow is, but mine is very straight forward production-mode testing that many people in my industry do. Even simple things like just running Figma in it's own native app runs slow, it's pathetic tbh.
To me, testing means real life scenarios and I can attest based on the big thread I made about planned obsolescence and Tahoe to me still rings true.
I can forgive Apple (a bit) that they don't care about Intel at this point because its EOL and Tahoe is the last supported OS version, however I cannot forgive them for the fact that they actually officially supported it with all these bugs. They should've dropped support by Tahoe if they barely care about it. I can also partially forgive them for not caring about Intel at all, because the M series chips are so much better (Like I said earlier, I have an M1 Max and M4 Pro here and can do side by side tests) and the M series feel significantly faster especially in single core performance. Obviously the GPUs don't even touch my 6900XT in terms of performance. So I am not hung on the fact that they want Intel users to move to M series chips, I am more hung up on the fact that they actually released a half baked OS and most likely it will never reach the performance of Sequoia.
On a surface level, on M series chips (especially M1 Pro and above) it runs OK, however, I am still seeing significant performance drops compared to Sequoia. The .3 Beta 3 issues are huge in Tahoe on M series chips, especially on the graphics side, but it's a beta so I forgive that.
You have to also understand that for example the M1 Max is a perfectly fine machine and if Apple is adding overhead to it, what is the point? Liguid Glass or WindowServer tweaks should not be something that slows down a system to push people to upgrade to the latest and greatest upcoming M5 Pro+ chips.
So if its "an extension" of some kind shouldn't there be some kind of paper trial in activity monitor? Wouldn't I be able to see some specific discrete point of consumption?
Depends where you’re seeing the slow downs. It could be a mutex being released way too late causing just a wait that doesn’t depend on or use system resources. Especially if slowdowns are localized to safari. Could be blocking the JavaScript thread or something
My M4 max with 1128 GB swaps on sequoia. Swap doesn’t mean you are short of memory. Linux and Mac OS aggressively swap and move unused apps/tabs to swap memory for faster reload or not swap when memory is really needed. I rarely restart my MBP, probably every few months.
Restart your computer at night and let it run with no apps overnight. Restart should clear the swap and any waits.
Restart your computer at night and let it run with no apps overnight. Restart should clear the swap and any waits.
Last edited:
My M1 Max with 64 GB is already upgraded. I have no issues, swap is same as before. I am testing and outside of compatibility of couple of libraries it’s not bad. I will upgrade my M4 max next month.
Do you not notice any lag at all? Do you use multiple spaces? I have 12 desktops.
And where do you notice lags?
My MacBook Air came with Sequoia and after starting it for the first time I directly upgraded to Tahoe. So is it possible that your installation is corrupted somewhere?