I am really torn between 8GB vs. 16GB RAM

[mo5214]

I found this : https://superuser.com/questions/97235/how-much-swap-is-a-given-mac-application-using

running this:
ps -o pid= -xa | awk '{print $1}' | xargs -n 1 sudo vmmap | egrep 'swapped_out|Path'

might yield some useful info about which processes are using swap.

 

[sbuntin]

That is an excellent point!

I've not seen upgrades add to Apple's trade-in values to any significant degree. If you're selling privately, upgrades will retain some value.

EDIT: After I wrote that, I thought I should doublecheck. Back in 2017, I purchased two almost identical MBP13's. One has 512GB, the other 256GB. I thought there was a memory upgrade too, but I don't have either machine handy to check. Difference in trade-in value? $30.

 

[Michael Scrip]

8GB vs 16GB for music production

 

[Ethosik]

But in this case. You can't just replace the SSD as it's soldered on, and paired to the system regardless of how the SSD failed. Personally I wished that Apple would go back to using replaceable SSDs. (I mean, they got that in the 2019 Mac Pros)

True. PCIe 4 NVMe should be the same speeds as Apple's SSDs

 

[OldMike]

Serious question, as I have read through all 1,004 posts so far in this thread.

In regards to swap and SSD wear, we can only guess at Apple's TBW rating for the drives it is using. I'm going to assume that it might be similar to the TBW rating of the Samsung 970 Pro, which would put the base 256GB SSD somewhere in the ballpark of 300TBW.

I've seen a lot of comments here that a few GB / day of SSD usage for swap is not a big deal. If I am thinking about this correctly, if your swap file is showing 4GB or 5GB - that is just its total size and not the amount that has actually been written to the file throughout the day. If the machine is low on memory and relying on swap, every time a browser tab is changed or application window is brought forward we could be talking about 400MB or 500MB swap used each time. On a constrained system this could really add up throughout the day.

Lets say I am working in multiple applications, and throughout the whole day switch 100 times between them for email, browser, slack, office apps, etc. Same with tabs on the browser, lets say throughout the day, on my memory constrained system, I switch tabs only 100 times.

If each application switch were to require an average of 500MB to be brought in and out of swap, that would mean throughout the day, your SSD writes would equal 200 x 500MB or 100 GB of data written from swapping alone. Actual swap usage may be even higher if more demand is being placed on the system. This does not count any normal use of OS and user file transfers to and from the SSD.

That would bring that internal SSD lifespan to 300TB / 100GB / 365 = 8.2 years without any actual use other than memory swap usage. Add onto that documents, photos, video, cached files from web, streaming, etc. being saved and deleted and an incredibly reduced lifespan seems like it would be a possibility to me.

Is my thinking wrong on this?

And wouldn't having an additional 8 GB physical memory help prevent many of these swaps throughout the day?

 

[rui no onna]

Serious question, as I have read through all 1,004 posts so far in this thread.

In regards to swap and SSD wear, we can only guess at Apple's TBW rating for the drives it is using. I'm going to assume that it might be similar to the TBW rating of the Samsung 970 Pro, which would put the base 256GB SSD somewhere in the ballpark of 300TBW.

I've seen a lot of comments here that a few GB / day of SSD usage for swap is not a big deal. If I am thinking about this correctly, if your swap file is showing 4GB or 5GB - that is just its total size and not the amount that has actually been written to the file throughout the day. If the machine is low on memory and relying on swap, every time a browser tab is changed or application window is brought forward we could be talking about 400MB or 500MB swap used each time. On a constrained system this could really add up throughout the day.

Lets say I am working in multiple applications, and throughout the whole day switch 100 times between them for email, browser, slack, office apps, etc. Same with tabs on the browser, lets say throughout the day, on my memory constrained system, I switch tabs only 100 times.

If each application switch were to require an average of 500MB to be brought in and out of swap, that would mean throughout the day, your SSD writes would equal 200 x 500MB or 100 GB of data written from swapping alone. Actual swap usage may be even higher if more demand is being placed on the system. This does not count any normal use of OS and user file transfers to and from the SSD.

That would bring that internal SSD lifespan to 300TB / 100GB / 365 = 8.2 years without any actual use other than memory swap usage. Add onto that documents, photos, video, cached files from web, streaming, etc. being saved and deleted and an incredibly reduced lifespan seems like it would be a possibility to me.

Is my thinking wrong on this?

And wouldn't having an additional 8 GB physical memory help prevent many of these swaps throughout the day?

Those estimates seem overly large. I've used SSDs on RAM-starved systems (well, Windows 7 32-bit so max ~3GB usable) and actual average daily writes is just ~20GB per day.

 

[ian87w]

I'm a bit confused with youtubers claiming that this unified RAM is more "efficient" and thus you don't need as much. Huh? RAM is RAM. The efficiency doesn't mean you suddenly need less RAM. It's like saying you can use a 512GB SSD to replace a 1TB HDD because the SSD is faster.... The speed and efficiency, to my understanding, has nothing todo with the amount. It's up to the OS to manage the RAM.

 

[onfire23]

Serious question, as I have read through all 1,004 posts so far in this thread.

In regards to swap and SSD wear, we can only guess at Apple's TBW rating for the drives it is using. I'm going to assume that it might be similar to the TBW rating of the Samsung 970 Pro, which would put the base 256GB SSD somewhere in the ballpark of 300TBW.

I've seen a lot of comments here that a few GB / day of SSD usage for swap is not a big deal. If I am thinking about this correctly, if your swap file is showing 4GB or 5GB - that is just its total size and not the amount that has actually been written to the file throughout the day. If the machine is low on memory and relying on swap, every time a browser tab is changed or application window is brought forward we could be talking about 400MB or 500MB swap used each time. On a constrained system this could really add up throughout the day.

Lets say I am working in multiple applications, and throughout the whole day switch 100 times between them for email, browser, slack, office apps, etc. Same with tabs on the browser, lets say throughout the day, on my memory constrained system, I switch tabs only 100 times.

If each application switch were to require an average of 500MB to be brought in and out of swap, that would mean throughout the day, your SSD writes would equal 200 x 500MB or 100 GB of data written from swapping alone. Actual swap usage may be even higher if more demand is being placed on the system. This does not count any normal use of OS and user file transfers to and from the SSD.

That would bring that internal SSD lifespan to 300TB / 100GB / 365 = 8.2 years without any actual use other than memory swap usage. Add onto that documents, photos, video, cached files from web, streaming, etc. being saved and deleted and an incredibly reduced lifespan seems like it would be a possibility to me.

Is my thinking wrong on this?

And wouldn't having an additional 8 GB physical memory help prevent many of these swaps throughout the day?

970 pro is a MLC drive. I am pretty sure apple is using a TLC drive like the 980 pro. The endurance is just 150TBW for the 256gb version

 

[turbineseaplane]

I'm a bit confused with youtubers claiming that this unified RAM is more "efficient" and thus you don't need as much. Huh? RAM is RAM. The efficiency doesn't mean you suddenly need less RAM. It's like saying you can use a 512GB SSD to replace a 1TB HDD because the SSD is faster.... The speed and efficiency, to my understanding, has nothing todo with the amount. It's up to the OS to manage the RAM.

I think they are referring to the way it’s swapping things around.

I also think most of them don’t know much and are just talking.

YouTube has turned every random yocal into a “reviewer” all the sudden...

I think many have dreams of being an MKBHD or the like.

It is truly stunning how much garbage and bad info is on YouTube

 

[Rck1984]

I bought a 8GB a few days ago and couldn't be more satisfied. That said, I am doing light work locally on the MacBook, and remotely connecting to a Windows server environment for the heavy work. Thus far, the 8GB is going strong!

I still think many people misunderstand how RAM works, reading things like "my RAM is nearly full, therefore I need more". RAM doesn't work like storage, where you fill up until there is nothing left.

Also, reading comments about future proofing. In a couple years, its not just the RAM that will hold your MacBook back. Your CPU, GPU and all the other stuff is most likely be outdated too (remember, M1 is in its first gen). I am a sucker for (new) tech, when this machine starts holding back I will gladly give it to my girlfriend, who doesn't get hurt over a tiny hiccup once in a while and get myself a new M2/M3.

 

[thinkingBanana]

Those estimates seem overly large. I've used SSDs on RAM-starved systems (well, Windows 7 32-bit so max ~3GB usable) and actual average daily writes is just ~20GB per day.

I am not sure if these estimates are that far away from the truth. I have measured my SSD write with smartmontool and see 1.2TB of SSD write when I simply coded in Xcode for 2-3 hours. Gnattu suggested that the unit conversion might be off, but let's say it is off by 10 times, I am still writing to my SSD 100GB+ in 3 hours.

Since my post is basically dead, I am hoping for gaining some attractions here and having someone telling me the data I got seems crazy and there is nothing to worry about.

 

[Thysanoptera]

I'm a bit confused with youtubers claiming that this unified RAM is more "efficient" and thus you don't need as much. Huh? RAM is RAM. The efficiency doesn't mean you suddenly need less RAM. It's like saying you can use a 512GB SSD to replace a 1TB HDD because the SSD is faster.... The speed and efficiency, to my understanding, has nothing todo with the amount. It's up to the OS to manage the RAM.

It is both faster and requires less RAM amount, it is ok to call it more 'efficient'. With unified memory every component of the SoC has access to the whole memory area, while with conventional approach you need to move stuff around.

With your SSD example, try to imagine that your OS only allows printing of files from /printer directory mounted on separate physical disk. If you want to print a picture from your Photo library, you need to move it (because storage space is precious) first to this /printer directory. To do that you need to check that there is enough space on disk, reserve this space, move your picture, print, then repeat the process to move the picture back to your photo lib. You will need exactly twice as much space and will spend much more time doing so than 'Unified OS' that allows you to print a file from any location on any disk.

 

[thinkingBanana]

Serious question, as I have read through all 1,004 posts so far in this thread.

In regards to swap and SSD wear, we can only guess at Apple's TBW rating for the drives it is using. I'm going to assume that it might be similar to the TBW rating of the Samsung 970 Pro, which would put the base 256GB SSD somewhere in the ballpark of 300TBW.

I've seen a lot of comments here that a few GB / day of SSD usage for swap is not a big deal. If I am thinking about this correctly, if your swap file is showing 4GB or 5GB - that is just its total size and not the amount that has actually been written to the file throughout the day. If the machine is low on memory and relying on swap, every time a browser tab is changed or application window is brought forward we could be talking about 400MB or 500MB swap used each time. On a constrained system this could really add up throughout the day.

Lets say I am working in multiple applications, and throughout the whole day switch 100 times between them for email, browser, slack, office apps, etc. Same with tabs on the browser, lets say throughout the day, on my memory constrained system, I switch tabs only 100 times.

If each application switch were to require an average of 500MB to be brought in and out of swap, that would mean throughout the day, your SSD writes would equal 200 x 500MB or 100 GB of data written from swapping alone. Actual swap usage may be even higher if more demand is being placed on the system. This does not count any normal use of OS and user file transfers to and from the SSD.

That would bring that internal SSD lifespan to 300TB / 100GB / 365 = 8.2 years without any actual use other than memory swap usage. Add onto that documents, photos, video, cached files from web, streaming, etc. being saved and deleted and an incredibly reduced lifespan seems like it would be a possibility to me.

Is my thinking wrong on this?

And wouldn't having an additional 8 GB physical memory help prevent many of these swaps throughout the day?

I think your calculation is reasonable, and the uncertainty is how accurate is the 100GB/day assumption, and I found that the endurance of the SSD is probably 150TBW instead of 300TBW. I did an SSD write measurement a while ago using smartmontool, and find that during IDLE my computer uses 30GB~, intense web browsing day using 230GB~, and 3 hours of Xcode with an emulator using 1.2TB~.

I would rather believe that these measurements are significantly off though; otherwise, I will be on track of grinding the SSD to the ground in less than 2 years.

What comforts me a little is the thought that I will get a more powerful Mac at that time anyway, and one report says SSD will last much longer than what manufacturers claimed.

edit: made it more readable

 

[OldMike]

970 pro is a MLC drive. I am pretty sure apple is using a TLC drive like the 980 pro. The endurance is just 150TBW for the 256gb version

I am not sure if these estimates are that far away from the truth. I have measured my SSD write with smartmontool and see 1.2TB of SSD write when I simply coded in Xcode for 2-3 hours. Gnattu suggested that the unit conversion might be off, but let's say it is off by 10 times, I am still writing to my SSD 100GB+ in 3 hours.

Since my post is basically dead, I am hoping for gaining some attractions here and having someone telling me the data I got seems crazy and there is nothing to worry about.

This would not seem like a good combination...

150TBW for a 256GB drive is a really disappointing rating if true. That is the same rating as the old Samsung 860 EVO TLC drives.

@thinkingBanana what are the specs of your Mac?

 

[thinkingBanana]

This would not seem like a good combination...

150TBW for a 256GB drive is a really disappointing rating if true. That is the same rating as the old Samsung 860 EVO TLC drives.

@thinkingBanana what are the specs of your Mac?

@OldMike I would really hope the rating is much higher as well... or the result I captured is grossly off.

And Oops, I should have listed what I have:

M1 Mac Mini; Base Config - 8GB RAM - 256GB SSD.

 

[OldMike]

@thinkingBanana I just read your other thread that you linked to.

I would like to get to the bottom of this also. I am thinking about the same config as you have for similar purposes, and would love to know the full story of the ssd drive swap stats you are seeing with your usage of Xcode.

The tool I use, DriveDX, does not appear to support the M1 natively and I am not sure if it would run in Rosetta. It is a free trial, though, if you want to give it a shot. I am sure there are plenty of other tools (maybe some free) that you can use to cross check the measurements the Smartmontool is giving you.

 

[JouniS]

It is both faster and requires less RAM amount, it is ok to call it more 'efficient'. With unified memory every component of the SoC has access to the whole memory area, while with conventional approach you need to move stuff around.

Unified memory is more efficient, because you don't have to move data around as much. On the other hand, you need more of it, because the GPU takes its own share. In normal desktop use, 8 GB of unified memory is equivalent to maybe 7 GB of RAM with a discrete GPU.

 

[thinkingBanana]

@thinkingBanana I just read your other thread that you linked to.

I would like to get to the bottom of this also. I am thinking about the same config as you have for similar purposes, and would love to know the full story of the ssd drive swap stats you are seeing with your usage of Xcode.

The tool I use, DriveDX, does not appear to support the M1 natively and I am not sure if it would run in Rosetta. It is a free trial, though, if you want to give it a shot. I am sure there are plenty of other tools (maybe some free) that you can use to cross check the measurements the Smartmontool is giving you.

@OldMike thanks for the tool suggestion, I have downloaded DriveDX and ran it, the result is attached.

Looks like the result is identical to smartmontool in terms of block count, but the same block count yields a different TB count. DriveDX reports 6.1TB write, while smartmontool reports 6.7TB write.

 

[armoured]

I'm a bit confused with youtubers claiming that this unified RAM is more "efficient" and thus you don't need as much. Huh? RAM is RAM. The efficiency doesn't mean you suddenly need less RAM. It's like saying you can use a 512GB SSD to replace a 1TB HDD because the SSD is faster.... The speed and efficiency, to my understanding, has nothing todo with the amount. It's up to the OS to manage the RAM.

Overall I agree, there's no magic. But to many users, the advantage of a faster system is going to 'feel like' the advantage of having an extra N gigabytes of ram. That's probably 90% of the difference, it feels like it has more memory because they have a mental model of how extra ram feels. (Think how an SSD made a system feel like a system with more memory in harddrive days).

Of course, if it does what they need - to those users, it really does 'mean' that they need less ram. But it's a subjective 'need' only.

The rest is perhaps some shorter-term spikes in ram needed overall perhaps falls a bit because of reduced collisions/overlaps in ram movements - but those short-term spikes might have a big impact on performance. (An analogy might be to a highway that gets dramatically worse if the on and offramps get jammed at rush hour - same quantity of cars, seems like you need a bigger highway).

Unified memory: I think it can result in less memory usage but not always, and depends what you're comparing to (a dGPU system with dedicated ram or ...?), with some programming specifics. But overall exaggerated and not magic; still, might make a difference from time to time.

There are theories out there about some advanced machine learning [something something] swap algorithms. Sounds made up to me - I think apple would more likely have touted such a change, if it existed. (But who knows, maybe it does and they're keeping it secret?) But compare: when apple introduced memory compression, they touted the advantages and said this meant some machines, sometimes, would need less memory; no secrecy, it was a selling point.

 

[ian87w]

Overall I agree, there's no magic. But to many users, the advantage of a faster system is going to 'feel like' the advantage of having an extra N gigabytes of ram. That's probably 90% of the difference, it feels like it has more memory because they have a mental model of how extra ram feels. (Think how an SSD made a system feel like a system with more memory in harddrive days).

Of course, if it does what they need - to those users, it really does 'mean' that they need less ram. But it's a subjective 'need' only.

The rest is perhaps some shorter-term spikes in ram needed overall perhaps falls a bit because of reduced collisions/overlaps in ram movements - but those short-term spikes might have a big impact on performance. (An analogy might be to a highway that gets dramatically worse if the on and offramps get jammed at rush hour - same quantity of cars, seems like you need a bigger highway).

Unified memory: I think it can result in less memory usage but not always, and depends what you're comparing to (a dGPU system with dedicated ram or ...?), with some programming specifics. But overall exaggerated and not magic; still, might make a difference from time to time.

There are theories out there about some advanced machine learning [something something] swap algorithms. Sounds made up to me - I think apple would more likely have touted such a change, if it existed. (But who knows, maybe it does and they're keeping it secret?) But compare: when apple introduced memory compression, they touted the advantages and said this meant some machines, sometimes, would need less memory; no secrecy, it was a selling point.

Good points. I agree that if Apple actually did something significant in macOS Big Sur in terms of memory management, they would make it into a tent pole point in WWDC, like what they did on memory compression. So it's safe to assume that Big Sur memory management is not much different than previous versions of macOS. Thus it's also safe to assume that memory requirements of apps is not changing significantly. The more responsive M1 system might put the illusion of the system having "more than enough," but I do still think that it is misleading on some youtubers claiming that 8GB RAM on M1 is "equivalent" to 16GB on intel. That makes no sense at all.

 

[ian87w]

I think they are referring to the way it’s swapping things around.

I also think most of them don’t know much and are just talking.

YouTube has turned every random yocal into a “reviewer” all the sudden...

I think many have dreams of being an MKBHD or the like.

It is truly stunning how much garbage and bad info is on YouTube

Yeah, too many tech youtubers that don't have a clue on the basics of computing hardware.

 

[armoured]

Good points. I agree that if Apple actually did something significant in macOS Big Sur in terms of memory management, they would make it into a tent pole point in WWDC, like what they did on memory compression. So it's safe to assume that Big Sur memory management is not much different than previous versions of macOS. Thus it's also safe to assume that memory requirements of apps is not changing significantly. The more responsive M1 system might put the illusion of the system having "more than enough," but I do still think that it is misleading on some youtubers claiming that 8GB RAM on M1 is "equivalent" to 16GB on intel. That makes no sense at all.

Yes, I agree, it doesnt' really make any technical sense.

But I'm a wee bit more sympathetic, only in the sense that if it's 'fast enough' for most, then the equivalence they're talking about is just a subjective 'feels like enough.'

Now, I think some will regret this, because eventually some will realise that the slowdowns are there, just less noticeable for now, and when they do more demanding stuff in future, it'll become more noticeable.

On the other hand, for some that day will never come, or far enough in the future that it won't matter to them, etc.

There's no one answer, just that some of the invocations of this as magical are overdone.

 

[alien3dx]

Yeah, too many tech youtubers that don't have a clue on the basics of computing hardware.

my best qoute in macrumor "ram not use no good". In old time we teach to reduce as small as possible now folks teach ram no use useless.

 

[fang-woem-rai]

I think we can make a pretty easy decision tree.

1. If your existing machine has 16GB or more RAM, then get 16GB.
2. If your existing machine has less than 16GB RAM, and you are wondering if you need 16GB, just get 8GB of RAM.

Imo people who actually needs the max RAM will know that they do and wouldn't be wondering about it.

I like the idea of decision tree in this case of selecting appropriate ram. But your tree is incorrect and doesn't take the important things into consideration like resale value, security, scope of use, non-upgradability, etc.. So i will modify it :-
1. If you just want to try these new M1 machines for a few weeks/months and don't plan to use it long term, get 8GB of ram.
2. If your existing machine has 16GB or more RAM, then get 16GB.
3. If your existing machine has less than 16GB RAM, and you are wondering if you need 16GB, just get 16GB of RAM.

Only those who are absolutely sure that they can get away with 8 GB of ram should get it in 2020/2021.

 

[darngooddesign]

Also, reading comments about future proofing. In a couple years, its not just the RAM that will hold your MacBook back. Your CPU, GPU and all the other stuff is most likely be outdated too (remember, M1 is in its first gen).

I'm not sure I understand this. How will the M1 hold the computer back when an equally fast Intel chip or M2 wouldn't? As more apps become native, they are going to run more, not less, efficiently.