RAM and Longevity

[mi7chy]

OS and software trend toward being more bloated over time. Plus, there's translation resource overhead if you use non-Apple Silicon native software. As inflated as Apple RAM upgrade prices are it's the lesser of two evils to pay +$200 difference now versus having to buy a new computer down the road if RAM becomes a bottleneck.

 

[henrikhelmers]

I do not understand what RAM has to do with longevity.

In days past it used to be a hard limit on what software you could run. As memory management has gotten better and hardware faster, the limit is less hard. But the mental model lives on.

 

[throAU]

In days past it used to be a hard limit on what software you could run. As memory management has gotten better and hardware faster, the limit is less hard. But the mental model lives on.

Download lmstudio and play.

We're on the cusp of RAM requirements exploding, depending on what you do with your machine.

What HAS happened in the past decade is that OS requirements did not expand as fast as memory capacities did. But LLMs and on-device "ai" features are going to increase ram consumption by a lot in the next 5 years. I mean right now, even modest size LLMs running on my M4 Max are consuming 8-16 GB of Ram just for the LLM.

 

[henrikhelmers]

Download lmstudio and play.

We're on the cusp of RAM requirements exploding

Time will tell. Would love to see hardware make leaps

 

[apostolosdt]

OP, it's fairly simple: [...]

I hope this is helpful.

OK, a textbook post!

 

[Alameda]

Hey everyone.
This question is just came to mind. Everywhere I read about Macs vs Windows says that Macs last forever. Yet at the same time, people are always complaining about not enough RAM. So is the RAM really that big of a problem? Are these people just average joes that don’t do enough to generally push base spec ram? It’s usually 2012-2016 Macs that they talk about still running just fine and speedy after all this time.

I think it’s because Apple Macs don’t have expandable RAM, and most Windows laptops and desktops let you easily add RAM. If you buy a 16 GB Windows machine and realize a year later that you need 32 GB, it’s not a problem. With a Mac, you need to replace the whole computer, unless you bought it with 32 GB to begin with.

 

[smirking]

In days past it used to be a hard limit on what software you could run. As memory management has gotten better and hardware faster, the limit is less hard. But the mental model lives on.

People also don't consider that everything involved in swap has gotten faster by orders of magnitude. The swap round trip used to be death, but consider how extreme that bottleneck was in comparison to what we have now. The RAM itself used to be a lot slower and would swap over a slower system bus through a slower drive interface and on through the ultimate choke point, a spinning platter drive.

Besides better management, everything is faster now. I have a 2009 MBP that I still use for special purposes. It was nearly unusable with a HDD, but once I swapped in an SSD, it became quite serviceable and it only has 8GB of RAM! I would actually use it for more things if there was more software support for such an old computer.

I mostly use it now to run a few old programs that I can't use on modern Macs.

 

[Andy_2341]

The only thing you can configure in a Mac these days is Ram, hence it divides opinion. Some saying you must have more, others not so much.

I’ve had a Mac last ten years and only upgraded cause I fancied a change, home computing only. So it can happen.

My 8gb M1 eats up everything I give it and asks for more. Renders 4K quickly, compiles code snappily, edits images smoothly. However plenty on this forum will looks at the computer processing stats and scream…. MORE.

I think I would be in the same place as you. A base spec M4 Mini should be enough because in all honesty I don't need a Mac. I can do everything except for a few very particular desktop things with my iPad. Those specific things-syncing an iPod Shuffle and ripping cds-are pretty much considered obsolete already and don't take much power.

 

[Allen_Wentz]

I do not understand what RAM has to do with longevity. The chance of hardware failure is independent of how much RAM one has. I suppose one can even make an argument that more RAM increases the chance of failure, since there are more failure points

Let me explain. It is not about hardware failure, it is about the fact that every year for 40 years now Mac OS and apps have taken advantage of more and more RAM. Always. E.g. the max RAM my 2016 MBP can take is 16 GB; the max RAM my M2 MBP can take is 96 GB; the max RAM an M4 MBP can take is 128 GB. The RAM progression has continued inexorably starting with the 128k of 1984.

My 2016 MBP rammed-out with it's 16 GB RAM, meaning my workflow became slower, less smooth and less consistently reliable circa 2020 or so, even though if anything I reduced my workflow load. Over time hardware and software engineers design to optimize using more RAM because RAM is a great way to compute, and we buy computers to compute with.

RAM is really fast, so having plenty of RAM can help an aging chip to continue to perform adequately farther into the future. Less RAM OTOH means paging to disk, slower operation, etc.

I strongly recommend reading up on Apple's Unified Memory Architecture.

 

[Boyd01]

I think it’s because Apple Macs don’t have expandable RAM,

The 2018 Mini (final Intel version) had expandable RAM. Apple only approved of having it done by an authorized dealer, but many people got the minimum and upgraded RAM themselves, although it involved a fair amount of disassembly.

Personally, I got a 64gb 2018 Mini from the Apple Refurb Store almost 5 years ago and have not regretted it.

 

[Allen_Wentz]

Thank you everyone for sharing your experiences. I'm left with two options it seems. Drop the extra 200 and go for 24GB of RAM in the Mini which should be fine for me till it isn't supported considering what I'm doing. Or take the base spec and trade it in should I run into problems. I'm leaning towards the second one since this will be my 1st Mac and I'm learning if I even like the OS.

Your identification of two options is spot on IMO. However when you suggest "I'm leaning towards the second one [take the base spec and trade it in should I run into problems] since this will be my 1st Mac and I'm learning if I even like the OS" you do need to be aware that lesser RAM puts a Mac into running suboptimally more often. So in that regard you will be learning with a non-optimal RAM-limited Mac.

The good news is that many users are oblivious to suboptimal operation because the Mac OS does a great job of managing suboptimal RAM by paging to SSD, and modern SSDs are pretty fast. Just be aware.

 

[Andy_2341]

Your identification of two options is spot on IMO. However when you suggest "I'm leaning towards the second one [take the base spec and trade it in should I run into problems] since this will be my 1st Mac and I'm learning if I even like the OS" you do need to be aware that lesser RAM puts a Mac into running suboptimally more often. So in that regard you will be learning with a non-optimal RAM-limited Mac.

The good news is that many users are oblivious to suboptimal operation because the Mac OS does a great job of managing suboptimal RAM by paging to SSD, and modern SSDs are pretty fast. Just be aware.

Thank you for the heads up. Maybe I should try a base spec and push it to the max of what I would ever plan to do and see if it works? If not, then I just use the return window. Can I return items purchased with education discount?

 

[leman]

Thank you for the heads up. Maybe I should try a base spec and push it to the max of what I would ever plan to do and see if it works? If not, then I just use the return window. Can I return items purchased with education discount?

That is indeed the best course of action. Apple can even extend the no return period if you call them. Everything else is idle speculation and urban myths.

Having more RAM extends the usable lifetime of the machine. I had a PC I used for 13 yrs because I could upgrade RAM then SSD then GPU. Still running in the closet via remote desktop when I need to run some Windows stuff

I never understood this type of argument. Sure, you can get more RAM, but you are stuck with the same slow CPU and same slow RAM interface. If your computational demands grow over time, or the newer OS is optimized for faster graphics or new CPU capabilities, more RAM won't help you. RAM is far from a panacea many PC users make it out to be.

My 2016 MBP rammed-out with it's 16 GB RAM, meaning my workflow became slower, less smooth and less consistently reliable circa 2020 or so, even though if anything I reduced my workflow load.

If your workflow became slower, that's certainly not because the lack RAM. More likely the slow hardware showing its age, especially on an OS that is designed with certain hardware capabilities in mind.

Over time hardware and software engineers design to optimize using more RAM because RAM is a great way to compute, and we buy computers to compute with.

This is simply not true. We (software engineers) definitely don't do that. Sure, over time there tends to be more bloat, because newer frameworks are less optimized than older ones, because graphical assets grow in size, and because inflation of RAM makes engineers do sloppy job. Sometimes you have a paradigm shift that can indeed increase the RAM demand, like retina displays in 2016 or large machine learning models. New standard for personal computing seems to be 16GB and unless the user has special needs, they will be fine with it for many years to come.

RAM is really fast, so having plenty of RAM can help an aging chip to continue to perform adequately farther into the future.

RAM does not make your computer faster. RAM can only ensure that your computer does not get slower if your working set increases. If your working set does not increase, you don't need more RAM.

I strongly recommend reading up on Apple's Unified Memory Architecture.

UMA has nothing to do with the topic.

 

[throAU]

RAM does not make your computer faster. RAM can only ensure that your computer does not get slower if your working set increases. If your working set does not increase, you don't need more RAM.

Caveat: "your" working set will increase as Apple adds background features to macOS. Because the total amount of RAM you need has to include both whatever work you're doing along with whatever macOS is doing.

If you keep the same version of macOS on the machine, sure - you'll probably be just fine to do the same things in the future. But people want continued software support and security updates.

macOS and applications get larger over time.

 

[leman]

Caveat: "your" working set will increase as Apple adds background features to macOS. Because the total amount of RAM you need has to include both whatever work you're doing along with whatever macOS is doing.

If you keep the same version of macOS on the machine, sure - you'll probably be just fine to do the same things in the future. But people want continued software support and security updates.

macOS and applications get larger over time.

That is true, although I'd like to understand better how much of this is attributed to the lack of RAM and how much to the overall lack of performance. Would be interesting to see a detailed empirical study on this.

 

[throAU]

That is true, although I'd like to understand better how much of this is attributed to the lack of RAM and how much to the overall lack of performance. Would be interesting to see a detailed empirical study on this.

These days I'd suggest for sure its more a case of inappropriate original spec, but there's a massive caveat right now:

• we're on the cusp of LLMs being run on device with Apple intelligence (amongst others running things like LMStudio, etc. locally).

Running local LLMs are one of the heaviest (in terms of ram) end user tasks I've encountered since I've been using computers (say, 1983?). And you don't need to be a power user for that to be the case, just loading a model can consume tens of GB of RAM, and the larger the model the better the potential results.


Outside of that, these days RAM capacities are still doubling occasionally, and that is outrunning the general pace of software development, outside of the LLM case above.


If you don't care about LLMs... fine - but you may need to manually disable them in macOS once apple starts making heavier use of them with future revisions of macOS 15 and later.

Maybe I'm wrong with the above assumption, and maybe Apple's baseline of 16 GB will be fine, but personally I wouldn't be thinking "oh neat, now I have 2x the RAM available vs. last year in the base model" - because I suspect that will be rapidly consumed by on-device processing in the near future (and that's why apple felt the need to bump the spec).

 

[leman]

These days I'd suggest for sure its more a case of inappropriate original spec, but there's a massive caveat right now:

• we're on the cusp of LLMs being run on device with Apple intelligence (amongst others running things like LMStudio, etc. locally).

That is very true. LLMs are one of those disruptive factors that change the equation. We had a similar spike in RAM usage due to high-res assets when HiDPI displays became popular.

 

[dmccloud]

These days I'd suggest for sure its more a case of inappropriate original spec, but there's a massive caveat right now:

• we're on the cusp of LLMs being run on device with Apple intelligence (amongst others running things like LMStudio, etc. locally).

Running local LLMs are one of the heaviest (in terms of ram) end user tasks I've encountered since I've been using computers (say, 1983?). And you don't need to be a power user for that to be the case, just loading a model can consume tens of GB of RAM, and the larger the model the better the potential results.


Outside of that, these days RAM capacities are still doubling occasionally, and that is outrunning the general pace of software development, outside of the LLM case above.


If you don't care about LLMs... fine - but you may need to manually disable them in macOS once apple starts making heavier use of them with future revisions of macOS 15 and later.

Maybe I'm wrong with the above assumption, and maybe Apple's baseline of 16 GB will be fine, but personally I wouldn't be thinking "oh neat, now I have 2x the RAM available vs. last year in the base model" - because I suspect that will be rapidly consumed by on-device processing in the near future (and that's why apple felt the need to bump the spec).

To me, this post places too much stock into LLMs becoming the future of computing. While Apple Intelligence is available on any M1 or later Mac, that is a far cry from running LLMs on the same machines. Most Mac users will never be running LLMs locally, and wouldn't know where to begin even if they had a big animated and talking arrow guiding them through the process.

The percentage of Mac users who will run LMStudio, Ollama, etc. is small enough that it doesn't even represent a plurality of the developer base, let alone the general userbase.

 

[leman]

To me, this post places too much stock into LLMs becoming the future of computing. While Apple Intelligence is available on any M1 or later Mac, that is a far cry from running LLMs on the same machines. Most Mac users will never be running LLMs locally, and wouldn't know where to begin even if they had a big animated and talking arrow guiding them through the process.

The percentage of Mac users who will run LMStudio, Ollama, etc. is small enough that it doesn't even represent a plurality of the developer base, let alone the general userbase.

I think the topic is Apple Intelligence. It does add to the active RAM usage.

 

[Alameda]

I think the topic is Apple Intelligence. It does add to the active RAM usage.

Not really. It uses RAM when you use it, not continually. I already tested that. So… don’t ask Siri to write a dissertation while you’re editing videos.

There are some people who have professional tasks that use a huge amount of RAM, but mostly, they don’t. On an 8 GB Mac, you can run Word, Excel, PowerPoint and a web browser all at once just fine. Sure, there are some Adderall addicts who “need” to have 100 browser tabs open at once, but normal people don’t need to do that.

When I’m editing photos, I shut down other applications and make sure I have plent of free disk space and my software screams. And when I’m doing ordinary office work, I have no problem at all with 16GB and I had no problem with 8GB before that, except when I had to run Linux under VMWare at the same time as my MacOS. I could still get everything done, but it was two 4GB operating systems, so I just had to keep things light, that’s all.

 

[throAU]

Will just drop this here…

Research

Explore advancements in state of the art machine learning research in speech and natural language, privacy, computer vision, health, and more.

machinelearning.apple.com

Apple intelligence is literally going to be running local models on device.

Not just LLMs but other AI models that will require a lot of GPU, NPU and RAM to hold the working set.

 

[AppaSquatic]

I do not understand what RAM has to do with longevity. The chance of hardware failure is independent of how much RAM one has. I suppose one can even make an argument that more RAM increases the chance of failure, since there are more failure points

Not sure, but I think it may have once had something to do with excessive wear of swap files of the SSD. But I think that's less of a problem with 16GB of RAM as base spec

 

[throAU]

Not sure, but I think it may have once had something to do with excessive wear of swap files of the SSD. But I think that's less of a problem with 16GB of RAM as base spec

RAM is to do with device viability.

• Over time, CPUs advance at between 5-30% performance improvement per year.
• However, if you start running out of of memory and into swap, performance doesn't degrade by 5, 10 or 30 percent, it rapidly totally falls off a cliff, like 2x, 5x, 10x slower if you're say, running 50% of your "hot data" out of swap rather than RAM.

This is why RAM is more important to long term device viability more so than CPU.

Put another way, 5 years of CPU improvements might net you a 2x performance increase (if you're lucky these days, M series has been great, intel has not), but half the RAM you need in 5 years time (due to being stingy on initial purchase) will get you a (for example) 2-5x performance decrease when it starts running short.

Depends very much what you're doing - but RAM requirements have only trended one way since the dawn of computing, and GENERALLY speaking, RAM and disk throughput requirements expand faster than CPU over the past 50 years or so.


ALL THAT SAID
Over capitalising on RAM is also stupid. RAM gets cheaper over time and 5 years into a machine (especially if its portable) you're looking at batteries that need replacing, ports that are well behind the times in terms of IO throughput, etc.

As always my advice is: plan on keeping the machine 3-5 years, and bump maybe ONE tier up on memory vs. what you need today. Don't overcommit because RAM gets cheaper over time, and there's no point having a machine with a huge memory capacity that is crap because everything else is out of date. You just way overpaid for memory you could buy cheap in 5 years time.

 

[AppaSquatic]

RAM is to do with device viability.

• Over time, CPUs advance at between 5-30% performance improvement per year.
• However, if you start running out of of memory and into swap, performance doesn't degrade by 5, 10 or 30 percent, it rapidly totally falls off a cliff, like 2x, 5x, 10x slower if you're say, running 50% of your "hot data" out of swap rather than RAM.

This is why RAM is more important to long term device viability more so than CPU.

Put another way, 5 years of CPU improvements might net you a 2x performance increase (if you're lucky these days, M series has been great, intel has not), but half the RAM you need in 5 years time (due to being stingy on initial purchase) will get you a (for example) 2-5x performance decrease when it starts running short.

Depends very much what you're doing - but RAM requirements have only trended one way since the dawn of computing, and GENERALLY speaking, RAM and disk throughput requirements expand faster than CPU over the past 50 years or so.

Makes sense

 

[Superman041]

OP, it's fairly simple:

1. Silicon Macs can't have any RAM upgrades. So,
2. Buyers have to try to anticipate how much RAM they will EVER need, not just how much they roughly know they need now.
3. No one really knows the answer to the "ever" question with 100% confidence, as Macs are usually good for upwards of 7 years and anything can happen in any 7-year span of time to anyone.
4. Meanwhile, Apple is the LONE "store" for Silicon RAM (and SSD) and they flex that "no competition" muscle in charging 3X-5X market rates over comparable RAM (and SSD).
5. Consumers are generally wanting to spend as little as they must, so not being able to anticipate hardware needs makes it difficult to choose. Wallet/budget wants "cheaper". Unknown future evolutionary needs want "more." There's no way to reconcile the two with complete confidence. So,
6. These variables create buying angst. We want to get it right because- if we don't- we can't fix it later with upgrades: it's essentially throw out the entire Mac and replace it with another entire Mac.
7. One more (fear) factor is called "SWAP", which is when a Mac runs out of free RAM and "borrows" a kind of virtual RAM by using the SSD like it is some extra RAM. This works... basically allowing too little RAM to be covered by SSD used as SWAP. HOWEVER, we all know that too many WRITES to an SSD is what wears them out. And when this internal SSD conks, the Mac is dead and has to be replaced. When one relies on SWAP to cover the RAM gap too often, we know that we are wearing out the SSD faster. This variable creates additional buyer angst.
8. Lastly, how much RAM is needed for core macOS use is not absolute either. Up until recently (up until Apple embraced 16GB as base), fans passionately argued that 8GB was enough for nearly everyone (and how convenient since that was exactly what Apple was pushing too by clinging to 8GB RAM). Of course, you see very little on that topic now, including any ripping into Apple for "forcing too much RAM" into every Mac sold today. Instead, now that Apple has shifted, so shifts fan opinion (as it always does). What's driving the shift? Most believe Apple embracing A.I. means that 8GB is NOT enough for 2025 and next few years as A.I. stuff in macOS keeps piling up and needs LOTS of RAM to function well.
9. Nobody outside of Apple knows how much RAM is needed by future versions of macOS, so we again must best guess, fueling more angst in trying to determine how much RAM we will need for macOS out in 2028-30 or later (however long we think we will want to use the Mac we purchase today).

Pull all this together and best plan is to try to anticipate your RAM needs for the LAST year you will own the Mac about to be purchased. And then configure to THAT need. For many, that means buying a Mac to use in about 2032 instead one that is "enough" in 2024-25... UNLESS, you are fine with regularly replacing "whole" Macs like we already regularly replace "whole" iPhones. Get this wrong now and that will be the fan advice: "just buy a new Mac."

For many, when they configure their new Mac with their best guess at covering all of the above, the fat Apple premium on RAM & SSD is very frustrating... especially if you have experience with past Macs offering the ability to buy "base specs" and then upgrading with third party RAM & storage at highly competitive prices instead of the fat Apple premium that helps drive "another record quarter" every quarter. So again, #5 steps in and we start trying to rationalize LESS RAM & SSD because we don't want to pay too much (when we can't even be sure about our computing needs out in those last years we'll use this Mac). Fru$$$$$$$tration!

To the Windows question
Boyd01 summarized it pretty well. ARM Windows emulation is not full Windows. If you need complete reliability, don't assume emulation will do the trick. If you barely need Windows and all you need of it is well known to work fine on ARM Windows PCs, you would probably be fine with emulation. However- and again during life of device, not just immediate needs- if you don't know about Windows app needs well into the future, the better plan is either to keep an old Intel Mac on which you can do your "full" Windows stuff via Bootcamp OR- do as I did- and buy yourself a PC too for "old fashioned bootcamp." There are Mac Mini-like PCs that are pretty loaded and the incredible competition in that world means you can buy a whole lot of PC for a Mac budget... and/or quite a good one for less than Mac budgets.

So, if you need "full" Windows and/or can't anticipate all Windows apps you'll want/need to run in the next 5+ years (and who really can?), you might serve your needs better to put some money towards buying a little PC too.

In my case, I chose a monitor with more than 1 input (so not ASD) and have both my Silicon Mac and a little PC sharing it. No problems (and no worry about) running ANY Windows app on the PC... AND it brings all of the benefits of a mountain of apps that run on Windows but are not available on Mac. For example, the PC world is actually serious about gaming, as evidenced by the abundance of desirable games available there.

IMO (summary advice)
For a Silicon Mac purchase, it is better to overbuy your best guess at RAM & SSD needs- even at Apple "exploitive" prices- than get out there in 2028 or 2030 and discover you "should've" and probably be replacing the entire Mac. Yes, it IS frustrating to compare what "more" RAM/SSD costs in a PC vs. Apple RAM/SSD but that's THE wallet-emptying burden if one wants enough Mac to probably cover all of their needs for life of device. No competition for anything ever benefits buyers: seller always charges much more with zero competitive pressure.

If your best guess at RAM and SSD needs out in the distant future implies you'll be using NORTH of maybe 80% of either or both, buy the next tier up above your best guess. #5 is quick to motivate us to underestimate (and rationalize it) but up to 20% or less "fudge factor" is NOT enough. Else, start saving for your next whole Mac... to likely be purchased sooner than you plan.

I hope this is helpful.

This is a great post and has add a bit more info to helping me decide on the post here as I trying to decide between the

1. M4 Pro 12-Core CPU , 16-Core GPU, 24GB Unified Memory , 512GB SSD Storage
2. Non M4 pro 10-Core CPU,10-Core GPU ,24GB Unified Memory 1TB SSD Storage .

I don't work heavily with video just yet, more social media video editing for now, multiple spreadsheets and soon to be coding (No LLM) . Confused because they're both the same price? According to ArtisRight, order of priority is the below. I want to future proof, I am someone who doesn't upgrade every 3/4 years. More like 7/8!

1. RAM
2. Chip family
3. SSD size
4. Chip varients


Thoughts?