News SSD capacity could quadruple by 2029 — 8Tb NAND will bring big and affordable SSDs to the market

[Admin]

3D NAND flash capacity will keep gaining layers and increasing storage.

SSD capacity could quadruple by 2029 — 8Tb NAND will bring big and affordable SSDs to the market : Read more

 

[Notton]

Affordable 8TB SSDs with decent endurance please. Gen3 or 4 speeds are fine.

 

[Stomx]

"We live in 2024. 2Tb QLC NAND memory devices are readily available..."

QLC = TIME BOMB made by tricky laws of planned obsolescence. Looks like worst nightmare of SSD manufacturers is SSD which never breaks

 

[Sippincider]

SSD capacity could quadruple by 2029

.... and a certain fruit vendor will still have 256gb base storage. 🤦‍♂️

 

[einheriar]

i think they should focuss more on affordable capacity now than gen 5 or gen6 speed.

 

[einheriar]

.... and a certain fruit vendor will still have 256gb base storage. 🤦‍♂️

and sell double the ssd capacity for $300 extra in a special budget offer

 

[Alvar "Miles" Udell]

With the arrival of 3D NAND about a decade ago, the storage capacity of solid-state drives has been increasing rapidly, and the per-GB price has been decreasing steadily.

Sadly with the arrival of PCIe 4.0 and now 5.0, the per-TB price has remained steady at around $100/TB, and 8TB SATA SSDs, for the consumer market anyway, are limited to the 3 year old Samsung 870 QVO which has doubled in price since October to over $600 again at Amazon, not to mention smaller format devices (such as cell phones) still bringing on the order of $100/128GB.

SSD capacity may quadruple, but I predict price per GB will remain relatively unchanged.

 

[Giroro]

QLC is trash. The slight increase in capacity was not worth the massive decrease in longevity and performance.
PLC will take a similarly huge performance hit, for even less uplift in capacity

 

[Alvar "Miles" Udell]

QLC is trash. The slight increase in capacity was not worth the massive decrease in longevity and performance.
PLC will take a similarly huge performance hit, for even less uplift in capacity

X-Nand was supposed to solve those issues, but nothing ever came from it.

https://www.tomshardware.com/news/x-nand-technology-gets-patented-qlc-flash-with-slc-speed

 

[usertests]

Sadly with the arrival of PCIe 4.0 and now 5.0, the per-TB price has remained steady at around $100/TB, and 8TB SATA SSDs, for the consumer market anyway, are limited to the 3 year old Samsung 870 QVO which has doubled in price since October to over $600 again at Amazon, not to mention smaller format devices (such as cell phones) still bringing on the order of $100/128GB.

SSD capacity may quadruple, but I predict price per GB will remain relatively unchanged.

Doom and gloom is warranted when it comes to SSDs and the chasing of more bits per cell. The technology is improving in density but not much else. Increased burst speeds from PCIe 5.0+ are irrelevant.

When it comes to pricing, I think we are in the bad part of a cycle. I remember 2 TB SSDs below $100 not long ago, particularly for budget brands like Team Group which got to around $60. Prices are way up from the low point but there are still decent deals around:

https://slickdeals.net/f/17794608-2...h-pcie-gen4-x4-nvme-m-2-ssd-100-free-shipping

It's believable that 8 TB SSDs could be $100 or less in the early 2030s, as the 1 Tb NAND chips get replaced by 4 Tb of the same size. But we're going to be stuck with at least QLC in consumer devices, guaranteed, and we've seen the industry talk of PLC (5 bits per cell) and beyond. It would be great if something fundamentally improved on NAND technology like that "X-NAND" you linked, or one of the many vaporware NAND killers over the years like Crossbar RRAM.

 

[TeamRed2024]

.... and a certain fruit vendor will still have 256gb base storage. 🤦‍♂️

and sell double the ssd capacity for $300 extra in a special budget offer

Ironically I'm watching an episode of Computer Chronicles from the 1980's and the topic is hardware. The same fruit vendor was selling less for more back then... and nothing has really changed over the last 40 years.

🤣

 

[Rookie_MIB]

Affordable 8TB SSDs with decent endurance please. Gen3 or 4 speeds are fine.

Oh geez. Yes. Please. I don't care about 'blazing 7GB/sec speeds', just give me a plain jane 8TB SATA 2.5" drive in the $200 range so I can build a NAS that doesn't have spinning rust storing my data.

Right now I'm limited to a 12TB NAS for my important stuff. I picked up a bunch of 2tb TeamGroup AX2's for a really good price ($65). At that rate, an 8TB SSD -should- cost somewhere in the $250 range. Nope, $600+ for an 8tb drive.

 

[TeamRed2024]

I don't care about 'blazing 7GB/sec speeds', just give me a plain jane 8TB SATA 2.5" drive in the $200 range so I can build a NAS that doesn't have spinning rust storing my data
an 8TB SSD -should- cost somewhere in the $250 range. Nope, $600+ for an 8tb drive.

I paid $399 each for my 2.5" 8TB Samsung 870 drives from Amazon over a year ago. I don't much care about transfer speed because all I use them for is digital media storage.

What's strange though is the exact same drive is showing on Amazon right now for $849.99 with a 27% discount down to $624... still a lot more than I paid.

Is there a parts shortage somewhere? I know my 4090 is currently retailing for more than I paid for it... but can't say I expected that to be the case with some 2.5" SSDs.

 

[USAFRet]

I paid $399 each for my 2.5" 8TB Samsung 870 drives from Amazon over a year ago. I don't much care about transfer speed because all I use them for is digital media storage.

What's strange though is the exact same drive is showing on Amazon right now for $849.99 with a 27% discount down to $624... still a lot more than I paid.

Is there a parts shortage somewhere? I know my 4090 is currently retailing for more than I paid for it... but can't say I expected that to be the case with some 2.5" SSDs.

Likely that is mostly advertising manipulation, just before the upcoming holiday "sales".

In another 4-6 weeks, you'll see that same drive at $450, (XX% OFF!!!), then down to $390 (XX% OFF!!)

 

[markhahn]

QLC is down to a few hundred cycles; PLC will be less than half that. Are we creeping up on WORM? It's mainly a question of how cold your data is...

Extra layers are bad scaling. Moore's Law is all about getting density for "free" (with a shrink). More layers is not free, it's linear in the number of layers. It takes twice as much effort to do 400 layers vs 200. (Yes, packaging is amortized, but it's not a serious component of device costs.)

 

[hotaru251]

"affordable" is missing asterisk.

we've seen it w/ ssd last yr & with ram multiple times. They will artificially keep prices "normal" rather than get cheaper by just reducing output should price get too low for their liking.

 

[sjkpublic]

Leave it to Shilov to post FLUFF.

 

[bit_user]

the number of bits stored per NAND memory cell has increased from two to three (triple-level cell, TLC) and four (quad-level cell, QLC)

In this case, you're doing a disservice by spelling these out and not at least indicating that it's actually the number of bits per cell. TLC packs 3 bits by using 8 levels, while QLC packs 4 bits by using 16 levels. Dumb names aren't your fault, but at least do your part to educate and not perpetuate the misunderstanding!

https://en.wikipedia.org/wiki/Multi-level_cell
​

@JarredWaltonGPU can you please ask Anton to be more mindful of this?

 

[Kamen Rider Blade]

QLC is down to a few hundred cycles; PLC will be less than half that. Are we creeping up on WORM? It's mainly a question of how cold your data is...

Extra layers are bad scaling. Moore's Law is all about getting density for "free" (with a shrink). More layers is not free, it's linear in the number of layers. It takes twice as much effort to do 400 layers vs 200. (Yes, packaging is amortized, but it's not a serious component of device costs.)

Spoiler: Endurance (Cycles) by Process Size & (Bit Levels/Cell)

Endurance Cycles were already bad with 3b/C.
It's only going to get worse with 4b/C.

I think we need to find ways to maintain that 1,000 Endurance Cycle Minimum Standard.
Otherwise, I wouldn't trust the SSD NAND Flash.

In this case, you're doing a disservice by spelling these out and not at least indicating that it's actually the number of bits per cell. TLC packs 3 bits by using 8 levels, while QLC packs 4 bits by using 16 levels. Dumb names aren't your fault, but at least do your part to educate and not perpetuate the misunderstanding!

​

​

https://en.wikipedia.org/wiki/Multi-level_cell​

​

@JarredWaltonGPU can you please ask Anton to be more mindful of this?

Can we as a industry go away from: ( <Insert Letter> Level Cell ) nomenclature?
Why can't we just use ( <#>b/C ) {e.g. 1b/C, 2b/C, 3b/C, 4b/C, 5b/C}?
Wouldn't just listing the actual # of bits per cell be FAR clearer to the reader and not play games?

I never liked MLC as 2-bits per cell, that was always confusing since it could conflate with it's later more dense breatheren.

DLC (Dual Levels per Cell) was supposed to be the nomenclature, but I don't see people or anybody in the industry using it.

Just listing the actual # should be clearer IMO.

The Prefix "Multi-" can be mean alot of different things in the English Language, and to the lay person, it would be confusing.

 

[newtechldtech]

by 2029 we should have new technology for SSD , not the same 3DNAND tech which was first introduced back in 2007.

 

[bit_user]

Spoiler: Endurance (Cycles) by Process Size & (Bit Levels/Cell)

Endurance Cycles were already bad with 3b/C.
It's only going to get worse with 4b/C.

You should cite your source on that. For one thing, I wonder how specific it is to a particular NAND vendor.

Can we as a industry go away from: ( <Insert Letter> Level Cell ) nomenclature?
Why can't we just use ( <#>b/C ) {e.g. 1b/C, 2b/C, 3b/C, 4b/C, 5b/C}?
Wouldn't just listing the actual # of bits per cell be FAR clearer to the reader and not play games?

Yeah, like I don't know why they can't just write 4bpc (bits per cell) instead of QLC.

I never liked MLC as 2-bits per cell, that was always confusing since it could conflate with it's later more dense breatheren.

I think it might have started out as a way to distinguish SLC from denser memories. However, when people started saying "triple-bit MLC", that somehow got shortened to TLC. Then, because TLC had become the established way of talking about 3-bits per cell, people came to associate "MLC" with 2 bits per cell. I can definitely see that happening. Doesn't make it okay, but at least it's understandable.

 

[jp7189]

Spoiler: Endurance (Cycles) by Process Size & (Bit Levels/Cell)

Endurance Cycles were already bad with 3b/C.
It's only going to get worse with 4b/C.

I think we need to find ways to maintain that 1,000 Endurance Cycle Minimum Standard.
Otherwise, I wouldn't trust the SSD NAND Flash.


Can we as a industry go away from: ( <Insert Letter> Level Cell ) nomenclature?
Why can't we just use ( <#>b/C ) {e.g. 1b/C, 2b/C, 3b/C, 4b/C, 5b/C}?
Wouldn't just listing the actual # of bits per cell be FAR clearer to the reader and not play games?

I never liked MLC as 2-bits per cell, that was always confusing since it could conflate with it's later more dense breatheren.

DLC (Dual Levels per Cell) was supposed to be the nomenclature, but I don't see people or anybody in the industry using it.

Just listing the actual # should be clearer IMO.

The Prefix "Multi-" can be mean alot of different things in the English Language, and to the lay person, it would be confusing.

Genuine question: how does endurance measured in total bytes written compare? Meaning how much does the capacity gain offset the endurance loss? E.g. a theoretical 3TB TLC vs a 4TB QLC which should have roughly the same number of cells.

 

[bit_user]

Genuine question: how does endurance measured in total bytes written compare? Meaning how much does the capacity gain offset the endurance loss?

That would be interesting, because if you just multiplied each bar by the number of bits per cell, then flash made on the 5X nm node would have more TBW capacity as 2-bit "MLC" than even in SLC mode. For some of the smaller nodes, however, it wouldn't be a net win.

Of course, there are additional variables when translating to TBW, which are things like write amplification and ECC.

 

[Kamen Rider Blade]

You should cite your source on that. For one thing, I wonder how specific it is to a particular NAND vendor.

I found that image too long ago, so I don't remember where I grabbed it off of.

 

[Hartemis]

Spoiler: Endurance (Cycles) by Process Size & (Bit Levels/Cell)

Endurance Cycles were already bad with 3b/C.
It's only going to get worse with 4b/C.

I think we need to find ways to maintain that 1,000 Endurance Cycle Minimum Standard.
Otherwise, I wouldn't trust the SSD NAND Flash.

Personally, I'd like to keep the data I store on a high-capacity ssd for all my life. I don't want to remove it and replace it with new data every day. So a few dozen write cycles, plus wear leveling, are more than enough for my use case.
I'm more concerned about offline retention. And QLC/PLC are bad for that too.