ssd swap - high usage of Terabytes Written

[leman]

The Why Apple's M1X Macs Don't Need 64GB of RAM! video kind of puts the kabosh on the "require gobs of RAM" thing that is the "mistake" in the something goes horribly wrong with efforts to built a PC video

As explained beginning at the 4:39 mark the unified memory is right there next to the CPU. On Intel machine it isi a ways away from the CPU and thanks to the lower bandwidth (2 channel vs the M1's 8 channel memory) a lot more has to be stored in the RAM to do the same thing ie the way Intel CPUs handle RAM is inefficient.

All this is wrong. Bandwidth and latency are identical between M1 and any modern x86 CPU running LPDDR4X RAM. They use the same amount of RAM channels and have same 128-bit memory interface. This is trivial confusion because LPDDR4X memory channels are 16bit where DDR4 channels are 64-bit. So 128-bit LPDDR4X memory configurations are reported as either 2, 4 or 8-channel RAM, depending on what you count.

This is why Intel CPUs "require gobs of RAM" "for an actual working set" - they have to hold things in memory because there isn't enough of a bandwidth between the CPU and the RAM and the distance that bandwidth has to travel.

This doesn't make any sense. Even if M1 had faster RAM (which it does not), how does faster RAM compensate for the need to store more things in RAM? If your working set is larger than the RAM, you still have to store the rest of the data somewhere. That "somewhere" is usually the SSD, so you have to constantly swap the data on and off the SSD. So the bandwidth of the RAM does not matter — the data fetch will be limited by the bandwidth of RAM-SSD interface.

I have to ask. Why do so many PC users have the idea that the solution to performance is largely solved by throwing more RAM at is rather than things like faster hard drives and the like? The idea has be coming from somewhere.

It's essentially a cargo cult effect. Back in the olden day (10-15 years ago) when RAM was expensive and PCs were shipped with meager 512MB or 1GB or RAM and everything tended to be RAM-starved, upgrading RAM could indeed be a huge boost (because in practical everyday terms, going from 512MB to 1GB is a much bigger difference than going from 16GB to 32GB — unless of course you are someone like @pshufd who actually needs it for their workflows). So the common recommendation was "upgrade your RAM", just like the more recent one is "get an SSD". Given that an average PC user is technically uneducated and every community needs it's traditions, these "upgrade tips" have stuck with us, even if the real world has moved on. It's really the same thing as "running the CPU hot will kill our computer", which was born from the overclocker community where one would mercilessly overvolt and burn out the CPUs.

 

[meropenem]

All this is wrong. Bandwidth and latency are identical between M1 and any modern x86 CPU running LPDDR4X RAM. They use the same amount of RAM channels and have same 128-bit memory interface. This is trivial confusion because LPDDR4X memory channels are 16bit where DDR4 channels are 64-bit. So 128-bit LPDDR4X memory configurations are reported as either 2, 4 or 8-channel RAM, depending on what you count.



This doesn't make any sense. Even if M1 had faster RAM (which it does not), how does faster RAM compensate for the need to store more things in RAM? If your working set is larger than the RAM, you still have to store the rest of the data somewhere. That "somewhere" is usually the SSD, so you have to constantly swap the data on and off the SSD. So the bandwidth of the RAM does not matter — the data fetch will be limited by the bandwidth of RAM-SSD interface.

I completely agree. I think the M1 Mac is great, but RAM is RAM, and you need a certain amount to be able to hold things in memory when you multitask. Making memory faster or decreasing latency does not change this.

Not sure why @Maximara needs to make the generalization that this desire for memory is from PC users wanting to throw more RAM at every problem. There's no harm in criticizing products that you otherwise enjoy.

 

[Maximara]

All this is wrong. Bandwidth and latency are identical between M1 and any modern x86 CPU running LPDDR4X RAM. They use the same amount of RAM channels and have same 128-bit memory interface. This is trivial confusion because LPDDR4X memory channels are 16bit where DDR4 channels are 64-bit. So 128-bit LPDDR4X memory configurations are reported as either 2, 4 or 8-channel RAM, depending on what you count.

"Besides the additional cores on the part of the CPUs and GPU, one main performance factor of the M1 that
differs from the A14 is the fact that's it's running on a 128-bit memory bus rather than the mobile 64-bit bus.
Across 8x 16-bit memory channels and at LPDDR4X-4266-class memory, this means the M1 hits a peak of
68.25GB/s memory bandwidth."

So what is this mobile 64-bit bus the article presented in the video referring too?

 

[leman]

So what is this mobile 64-bit bus the article presented in the video referring too?

The iPhone/iPad (A14)

 

[vs40]

After exact 1 month of usage:

Data Units Read:                    2,467,839 [1,26 TB]
Data Units Written:                 1,980,843 [1,01 TB]

It is fine with me.

 

[NoPlansForThis]

Well, just to update my case, Apple changed the entire motherboard.

All it took was a lot of phone calls with the Genius Bar where they tried to explain to me that "SSDs do not have limits to their lifespan! They can write data indefinitely! There is nothing wrong with 120tb written in 3 months!" The technician claimed that TBW was not an existing metric in the industry and I mentioned data sheets from manufacturers.

After I asked for this guarantee of "your SSD will live forever behaving like this" in writing, they changed my board.

The new one still writes a lot. I decided to give Monterey Beta a try to see if it makes any difference.

Tbh this kind of sounds like a bad Ram chip. I’ve experienced this a couple of times over the years. Same result though since Ram is also soldered. You’re probably getting a new logic board my man.

This is a real possibility, I will monitor the new one too

But its milady 😄

Sorry mang I gotta call BS on this. It is not reasonable to expect bad ssd longevity due to a cpu architecture change. Some apps won’t work well or at all, some stuff is unoptimized, some more crashes than usual, reasonable assumptions. But how could anyone know the ssd would get hammered and in what way does the architecture being new make this acceptable or something that could be reasonably expected?

A thousand times this. Just quoted so more people can read it, or re-read it.

The point is that the people who try to defend Apple by claiming a $1,000 M1 laptop isn't up to the task of... web surfing... make the Optiplex sound much better.

Apple markets this machine for machine learning but web surfing or using chrome in any capacity are too much. And the blame, apparently, should be directed to the user that needs ServiceNow, Zoom, Chrome. Not to the architecture that allows a simple task to destroy the hardware.

My two cents, after five months my SSD is almost a virgin.

I never log out or restart, my MBA 16/1TB is on 24/7. Zoom is not a problem. Happy customer here.

This makes me double down on my theory that the problem is a combination of hardware issues (maybe one specific chip manufacturer or batches) and Big Sur insane swapping behavior.

It's absolutely a problem that needs addressing. Not affecting everybody doesn't mean it's not a significant problem. 11.4 made significant improvements, but it's still not good enough. I have to go through quite a bit of rigamarole to keep my SSD writes donw to acceptable on a base model M1 air. There are certain programs/conditions which cause basically a kernel excessive write condition where it's needlessly writing and writing in the background, and I can't have powernap=1 without waking up to 100gb written while the computer is "sleep" mode. That's not acceptable long term.

Im testing Monterey to see if anything changed on that front, but have little hope. From my interaction with Apple, the standard answer is "there is no problem and SSD writes cannot degrade the lifespan of the disk"

If you have excessive writes, then diagnose the problem. Try your programs alone, one by one, and find the programs that result in excessive writes. I think that we've determined that this is NOT a widespread problem. Since the vast majority can't reproduce the problem, then someone who can reproduce it either has to demonstrate it to someone who can do something about it or the individual customer has to do some of the diagnostic work to show to Apple.

I have brought crash dumps to the Genius Bar to show them the component of the device having the problem.

What do you expect those of us that can't reproduce the problem to do?

Well, good luck diagnosing the problem, with or without Apple. From my first interaction with them I repeatedly asked for instructions to log my machine activity and provide data for them. I was dismissed every time. "There is no need for that, this would only be useful to the engineering team and we certainly will fix your machine before needing to involve them".

I had to try and diagnose it myself using smartmontools, DriveDX, whateaver info I could from Console and Activity Monitor, and kernel crashes. And Apple technicians still refused to even receive my data.

Just to add insult to injury, my complain was "whenever I am on a video call for too long - 1h or more - and sharing my screen, cpu usage and ssd writing go overboard on any of the apps. The machine freezes and crashes soon afterwards"

Their attempt to reproduce my problem was to open a "8k video" on YouTube and play it for a few hours on Safari and run Heaven benchmark.

I had little to no support here too besides " tell your IT people to chose other apps and quit zoom".

 

[CMMChris]

This is how it looks for me at the moment. Still all fine.

For those that don't remember: I never faced the SSD trashing issue. Release machine in daily use since late November 2020. It replaced my Hackintosh (32GB RAM, i7-8700K, Radeon VII) that I maxed out regularly. I also max this one out regularly, especially during 8k60 editing in Final Cut. Since 11.4 "Data Units Written" is growing slower due to less swapping as per my observation. Previously, swap was used even if memory pressure wasn't that high. Now it only uses swap if memory pressure gets into the orange area.

 

[Thistle41]

Well, just to update my case, Apple changed the entire motherboard.

All it took was a lot of phone calls with the Genius Bar where they tried to explain to me that "SSDs do not have limits to their lifespan! They can write data indefinitely! There is nothing wrong with 120tb written in 3 months!" The technician claimed that TBW was not an existing metric in the industry and I mentioned data sheets from manufacturers.

After I asked for this guarantee of "your SSD will live forever behaving like this" in writing, they changed my board.

The new one still writes a lot. I decided to give Monterey Beta a try to see if it makes any difference.




This is a real possibility, I will monitor the new one too

But its milady 😄



A thousand times this. Just quoted so more people can read it, or re-read it.



Apple markets this machine for machine learning but web surfing or using chrome in any capacity are too much. And the blame, apparently, should be directed to the user that needs ServiceNow, Zoom, Chrome. Not to the architecture that allows a simple task to destroy the hardware.



This makes me double down on my theory that the problem is a combination of hardware issues (maybe one specific chip manufacturer or batches) and Big Sur insane swapping behavior.



Im testing Monterey to see if anything changed on that front, but have little hope. From my interaction with Apple, the standard answer is "there is no problem and SSD writes cannot degrade the lifespan of the disk"



Well, good luck diagnosing the problem, with or without Apple. From my first interaction with them I repeatedly asked for instructions to log my machine activity and provide data for them. I was dismissed every time. "There is no need for that, this would only be useful to the engineering team and we certainly will fix your machine before needing to involve them".

I had to try and diagnose it myself using smartmontools, DriveDX, whateaver info I could from Console and Activity Monitor, and kernel crashes. And Apple technicians still refused to even receive my data.

Just to add insult to injury, my complain was "whenever I am on a video call for too long - 1h or more - and sharing my screen, cpu usage and ssd writing go overboard on any of the apps. The machine freezes and crashes soon afterwards"

Their attempt to reproduce my problem was to open a "8k video" on YouTube and play it for a few hours on Safari and run Heaven benchmark.

I had little to no support here too besides " tell your IT people to chose other apps and quit zoom".

Well, thanks for the update. I'm still getting much more TBW than others on here and I for one am still concerned about the potential impact on the SSD lifespan.

Here is my latest log:

16/07/2021	96,931,715	49.60		Logout/in	Removed whole Macos disk from TM
17/07/2021	97,107,583	49.70		Logout/in
18/07/2021	97,507,781	49.90		?
19/07/2021	98,137,278	50.20		?
20/07/2021	99,130,295	50.70		sleep only
21/07/2021	101,984,269	52.20		sleep only
22/07/2021	105,292,928	53.90		sleep only + charging
23/07/2021	105,427,457	53.90		logout/in
24/07/2021	105,437,355	53.90		logout/in
25/07/2021	105,491,531	54.00		sleep only
26/07/2021	106,041,737	54.20		sleep+charging

Strangely, it looked for a while that keeping it on charge when in sleep mode helped a lot. Logout/in definitely helps so I will follow that for now and it's not a great pain to wait a minute or so for things to settle.

I am banking on Monterey delivering the goods so I will be interested in your results thanks. I'm not keen on installing the Beta as I have a viable workaround for now.

 

[Thistle41]

This is how it looks for me at the moment. Still all fine.
View attachment 1811085
For those that don't remember: I never faced the SSD trashing issue. Release machine in daily use since late November 2020. It replaced my Hackintosh (32GB RAM, i7-8700K, Radeon VII) that I maxed out regularly. I also max this one out regularly, especially during 8k60 editing in Final Cut. Since 11.4 "Data Units Written" is growing slower due to less swapping as per my observation. Previously, swap was used even if memory pressure wasn't that high. Now it only uses swap if memory pressure gets into the orange area.

So, it looks like having the 16Gb RAM option avoids the SSD excessive writes. I'm up to 54 TBW in 6 months. I always install the latest official release do all those tweaks at the head of this thread which do help but not enough.

 

[CMMChris]

No it does not, there are plenty of 16GB users affected by the issue.

 

[jdb8167]

So, it looks like having the 16Gb RAM option avoids the SSD excessive writes. I'm up to 54 TBW in 6 months. I always install the latest official release do all those tweaks at the head of this thread which do help but not enough.

54 TBW in 6 months is probably not a problem. That should last for well over 10 years.

 

[pshufd]

One solution would be to get two Mac minis or a spare system and then split the workload between them - instant 32 GB M1X.

 

[Maximara]

Well, just to update my case, Apple changed the entire motherboard.

All it took was a lot of phone calls with the Genius Bar where they tried to explain to me that "SSDs do not have limits to their lifespan! They can write data indefinitely! There is nothing wrong with 120tb written in 3 months!" The technician claimed that TBW was not an existing metric in the industry and I mentioned data sheets from manufacturers.

Say what?! Do they actually believe people are that ignorant and/or stupid? Or worse, are they that ignorant and/or stupid?

 

[osplo]

I had to try and diagnose it myself using smartmontools, DriveDX, whateaver info I could from Console and Activity Monitor, and kernel crashes. And Apple technicians still refused to even receive my data.

Just to add insult to injury, my complain was "whenever I am on a video call for too long - 1h or more - and sharing my screen, cpu usage and ssd writing go overboard on any of the apps. The machine freezes and crashes soon afterwards"

Hey, "kernel crashes" and "machine freezes" are a completely different symptom from what I see other posters were reporting. You maybe right and maybe what you have is a compound effect of things involving a hardware / software / configuration failure. Let's see if the board change is useful. Please report.

Also, did you migrated to M1 using a clean install or you used a time machine backup? Maybe there was some kind of cascading trouble in the restoring process. In my case (I have no problem at all with SSD writes) I started from a clean slate.

 

[dogslobber]

So I'm averaging about 20GB a day with 11.5 and I'm thinking we can officially close this thread as problem solved. This is from an M1 MBA 8GB.

 

[Thistle41]

So I'm averaging about 20GB a day with 11.5 and I'm thinking we can officially close this thread as problem solved. This is from an M1 MBA 8GB.

So, how long have you had this update installed, please? I only did this yesterday so too soon to determine but overnight it was 100GB. Not too bad but not what I would hope for. We will see how it pans out over the next few days.

 

[dogslobber]

So, how long have you had this update installed, please? I only did this yesterday so too soon to determine but overnight it was 100GB. Not too bad but not what I would hope for. We will see how it pans out over the next few days.

5-6 days I think. I used to be utterly negative and disabled spotlight indexing due to this in earlier 11.x releases.

Now the kinks have been worked out, I think Apple may be onto something with these ARM based systems.

 

[Queen6]

Only spike in writes to the SSD I've observed is when I unlatched the AV to validate all files as opposed to only new ones. Adding 10GB which is understandable as the app has to decompress files to interrogate them.

Those seeing TB of writes daily should be speaking with Apple as that's not the norm. Much of my applications are open source and run under Rosetta with zero issue. As for Apple being on to something, absolutely yes as my M1 MBP easily sees of this fast hex core notebook with likely a binned down CPU being 3rd/4th fastest documented holding a solid 3.9GHz across all core irrespective of load.

Q-6

 

[Maximara]

Those seeing TB of writes daily should be speaking with Apple as that's not the norm. Much of my applications are open source and run under Rosetta with zero issue. As for Apple being on to something, absolutely yes as my M1 MBP easily sees of this fast hex core notebook with likely a binned down CPU being 3rd/4th fastest documented holding a solid 3.9GHz across all core irrespective of load.

The problem with talking with Apple is there may be little they can do (or you get some idiot who doesn't want to admit there is a problems and claims SSD wear doesn't happen). The reason for that is this isn't happening to everyone and no one really knows why it happens to some people and not others. IMHO you better off trying to figure out yourself as odds are it is likely some piece of third party software doing something stupid.

 

[Queen6]

The problem with talking with Apple is there may be little they can do (or you get some idiot who doesn't want to admit there is a problems and claims SSD wear doesn't happen). The reason for that is this isn't happening to everyone and no one really knows why it happens to some people and not others. IMHO you better off trying to figure out yourself as odds are it is likely some piece of third party software doing something stupid.

Maybe a better way to put it is; the more that raise the issue with Apple the greater the chance it will be investigated. I Don't see the issue myself and tend to agree it's likely down to third party SW or we would all be in the same boat so to speak.

Also has to be said that not all dev's explicitly follow Apple's guideline's, nor likely to come forward if their app is misbehaving. I looked at my own M1 MBP in some detail and far as I can ascertain it's as expected, with nothing out of the ordinary.

Q-6

 

[Maximara]

Maybe a better way to put it is; the more that raise the issue with Apple the greater the chance it will be investigated. I Don't see the issue myself and tend to agree it's likely down to third party SW or we would all be in the same boat so to speak.

But if the reports given here are any guide how would Apple investigate? We have dozens of reports that amount to "there is a problem" but no freaking details. Next to nothing about the programs being run, rarely is the RAM stated, a great number of go off raw numbers not the percentage used (and when we can cross check the two we tend to get what amounts to whining about 1 million when your budget is trillions)

Also has to be said that not all dev's explicitly follow Apple's guideline's, nor likely to come forward if their app is misbehaving. I looked at my own M1 MBP in some detail and far as I can ascertain it's as expected, with nothing out of the ordinary.

Given the kind of crap Apple has had with stealth code (Epic and others) how would Apple even know which programs are cauing the issues epscially if the problems is buried so deep it only shows up after the programs is in use for a hour or more?

 

[pshufd]

But if the reports given here are any guide how would Apple investigate? We have dozens of reports that amount to "there is a problem" but no freaking details. Next to nothing about the programs being run, rarely is the RAM stated, a great number of go off raw numbers not the percentage used (and when we can cross check the two we tend to get what amounts to whining about 1 million when your budget is trillions)

Given the kind of crap Apple has had with stealth code (Epic and others) how would Apple even know which programs are cauing the issues epscially if the problems is buried so deep it only shows up after the programs is in use for a hour or more?

I used to work in corporate support (we had regions report to us when they ran into a problem that they couldn't solve). We ideally wanted a reproducer which we could just run on our own systems but sometimes you get something that's intermittent or even hard to reproduce. So we asked for crash logs if there were crashes involved, or logs for when the problem is seen or we'd create an instrumented image to try to diagnose the problem. Sometimes we flew out to the customer for a couple of days or longer to sit with them to see the problem as it occurred. This was back in the day when companies paid a lot of money for support.

 

[Queen6]

But if the reports given here are any guide how would Apple investigate? We have dozens of reports that amount to "there is a problem" but no freaking details. Next to nothing about the programs being run, rarely is the RAM stated, a great number of go off raw numbers not the percentage used (and when we can cross check the two we tend to get what amounts to whining about 1 million when your budget is trillions)

Given the kind of crap Apple has had with stealth code (Epic and others) how would Apple even know which programs are cauing the issues epscially if the problems is buried so deep it only shows up after the programs is in use for a hour or more?

All of which is applied more here as people expect help with little if anything detailed documented. Apple can certainly dig through it's telemetry, system logs and see if it's directly related to the OS. Bottom line is if people don't provide the feedback Apple is unlikely to react and focus other issues.

I doubt it would be overly difficult isolate the cause if a third party application was responsible, equally resolving the issue another matter.

Q-6

 

[pshufd]

All of which is applied more here as people expect help with little if anything detailed documented. Apple can certainly dig through it's telemetry, system logs and see if it's directly related to the OS. Bottom line is if people don't provide the feedback Apple is unlikely to react and focus other issues.

I doubt it would be overly difficult isolate the cause if a third party application was responsible, equally resolving the issue another matter.

Q-6

You just need someone with good diagnostic skills to check a bunch of things. These people do tend to be expensive though.

 

[Earl Urley]

Hate to mention this, but I replaced Apple's SSD with an NVMe almost 2 years ago on this older 2014 laptop and I've only recently upped it from Mojave to Catalina, and in all that time its managed to accumulate 28.5 TB in writes according to DriveDX. It's used as a daily driver machine that's constantly running Zoom and downloading updates / PDFs / videos from iTunes..