Hacker News Reader

Comment by joe_mamba | Hacker News Reader

Man, I do wonder what the realistic lifespan of that single NAND chip will be after it gets hammered by constant swapping of running tasks way beyond the capabilities of a 8GB RAM machine.

I have a PC with a 10+ year old 256GB SATA Samsung SSD that's still in top shape, but that's different because that drive has those 256GB split over several NAND chips inside, so wear is spread out and shuffled around by the controller to extend lifespan. But when your entire wearable storage is a single soldered chip, I'm not very optimistic about long term reliability.

View on HN

There was quite a bit of discussion about that when the M1 first came out, but none of it really seemed to have happened six years later. The target audience isn't in danger of wearing it out and the ones that will push the limits will grow tired of it and sell it in a year or two or move on to the Neo 2, which might have 12gb of ram due to the expected chip.

I still think it's a great machine, but I think all these worries about NAND dying really haven't come to fruition, and probably won't. I have about a hundred plus of various SSD Macs in service and not one has failed in any circumstance aside from a couple of battery issues (never charged and sat in the box for 2 years, and never off the charger).

>There was quite a bit of discussion about that when the M1 first came out, but none of it really seemed to have happened six years later.

1. How do you know nothing happened? Define nothing in this case. Do Mac users check and report their SSD wear anywhere?

2. Didn't the OG 256gb M1 have 2 128MB NAND chips instead of one 256 meaning better wear resistance?

If swapping was causing SSDs to fail on M1 Macs, we would never see the end of the hysterical articles about "NANDgate". Since we haven't seen any in all these years, it's seems pretty certain it's not happening.

Hysteria would be if all had an issue like the keyboard gate, but this isn't an issue, it's a design limitation for certain uses cases which not everyone has. Some users will wear out faster than others due to usage patterns. If their M1 dies after 6 years of heavy usage, do you think they'll investigate if it was the NAND that died and go online to tell the news, or will they chuck it and buy new one?

NAND is still the same wearable part that regular X64 laptops have, Apple doesn't use some magic industrial grade parts but same dies that Samsung, Micron and SK ship to X64 OEMS, and those are replaceable for a reason, because they eventually fail.

The reality is most 8GB M1 Macs are still working just fine 6 years later. Power users know they need more than 8GB of RAM and will buy a MacBook Air or Pro with 16GB+.

The MacBook neo is for students, grandparents, travel, etc.

Hell, even if it dies after 6 years it was still a better experience than using a $500-600 windows PC and the cost comes out to ~$8/month spread over 6 years.

>The reality is most 8GB M1 Macs are still working just fine 6 years later.

Do you think SSD drives are replaceable for no reason? Just because M1 mac aren't failing left and right doesn't mean their NAND won't fail.

Even though I like the NEO, I can't in good faith buy a machine with soldered wearable parts. That's like buying a car with soldered brake pads because "in 6 years average users don't feel like they need changing".

I still had laptops on my hands from 20 years ago that work fine simply because you can swap their drives with fresh ones. How many M1 mac will still be functional in 20 years?

"How many M1 mac will still be functional in 20 years?"

Probably quite a few, MacBooks have had soldered SSD's for over 10 years now. My 2018 McBook Pro still has a perfectly functioning SSD. I still see people using 2015 and older MacBooks all the time. There is no widespread SSD failure issue after 10+ years of Apple soldering the SSD's.

For most people the SSD's are lasting longer than the useful life of the device.

Just because it's soldered doesn't mean it can't be replaced.

(But it's encrypted, so you'd better have backups because you can't read it off the chips.)

> Do you think SSD drives are replaceable for no reason?

The number one reason why laptop OEMs primarily use replaceable SSDs is so that they can switch SSD vendors on a monthly basis to whoever is the lowest bidder at the moment. The number two reason is so that they can offer multiple storage capacity options without building different motherboard configs (though in practice, a lot of OEMs never get around to actually selling the alternative configs). Repairability is a very distant third place.

Exactly. If some sort of random Dell model has a failure, you'll never hear about it because there's only a few thousand or so in circulation. But if any Apple product which sells in the tens/hundreds of millions has an issue, you'll hear about it whether you want to or not.

> Do Mac users check and report their SSD wear anywhere?

As a data point: I got a 14" MacBook Pro with a 512 GB SSD the first day it was available in 2021, and I've used it daily since then.

According to the SMART data ("smartctl -x /dev/disk0"), the SSD "percentage used" is 7%, with ~200 TBW. At this rate, the laptop will probably outlive me.

>but that's different because that drive has those 256GB split over several NAND chips inside, so wear is spread out and shuffled around by the controller to extend lifespan. But when your entire wearable storage is a single soldered chip, I'm not very optimistic about long term reliability.

I thought wear leveling worked at the page/block level, not the chip level? On an SSD, if there was a failure of an entire chip, you're still screwed.

You're correct, GP's understanding of how wear leveling works is off by several layers. Counting the number of BGA packages tells you nothing. There are multiple NAND dies per package, multiple planes per die, many blocks per plane, and the size of each erase block is the largest-scale feature that is relevant to wear leveling.

While high-density NAND is definitely worrying from a data retention standpoint, your reasons don't make sense. There has been a decrease in reliability with higher densities, and unless Apple is using SLC (strong doubt) you would expect around the same as any other manufacturer.

The sibling comments mentioning endurance don't tell the complete story either; continuously writing a drive until it shows errors means the cells have become leaky enough that they can't even hold data between each write and verify pass (hours or minutes apart), and while people point to such studies as "proof" that NAND endurance isn't something to worry about, they forget that endurance and retention are inversely related, as with temperature, and this is a statistical effect, so the true specification is more like "X years/months at temperature T after Y cycles with a BER of Z"; each one of those variables can be adjusted to make the others look as good or bad as you want.

from what i seen in "low end" ssds like the "120gb sata sandisk ones" under windows in heavy near constant pagging loads is that they exceed by quite a lot their manufacturer lifetime TBW before actually actually started producing actual filesystem errors.

I can see this could be a weaker spot in the durability of this device, but certainly it still could take a few years of abuse before anything breaks.

an outdated study (2015) but inline with the "low end ssds" i mentioned.

https://techreport.com/review/the-ssd-endurance-experiment-t...

The M2 MacBook Air base model has 8GB RAM and a single 256GB NAND chip. Nearly 4 years later, it doesn't seem to have caused any problems.

Most flash has average wear out after 300k cycles. Let's say 64GB is used for swap. That's 19200 TB or 19.2 PETABYTES of Swap usage. Let's say you swap 12GB a day, you will burn out that 64GB of Flash Storage in 4.38 years and my guess is that amount of swap usage is extremely high that user would probably replace laptop sooner out of performance frustration.

that doesn't maths

12GB a day isn't very much. If your working set is larger than the 8GB RAM, you're swapping multiple times per second. It doesn't take very many megabytes per swap to reach 12GB if you're doing that multiple times per second.

>Most flash has average wear out after 300k cycles

No it doesn't. Most 1TB drives are rated for around 600 TBW, so enough to overwrite the drive 600 times, nowhere near 300k cycles. If you search for specs of NAND chips used in SSDs, you'll find they're rated for cycles on the order of hundreds to thousands, still nowhere near "300k".

https://www.techpowerup.com/ssd-specs/crucial-mx500-4-tb.d95...

Original mid-90s Toshiba "solid state floppy disk" SLC flash: 1M cycles

2000s SLC flash: 100K cycles

Modern SLC/pSLC flash: 30-60K cycles

2010s MLC flash: 5-10K cycles

Modern QLC flash: 300-500 cycles

...and I won't even get into the details of their retention characteristics, suffice to say they subtly redefined them over the years to make the newer numbers better than they really are.

Reply on news.ycombinator.com