Why do SSD's have a large jump in capacity @ 4TiB+?

[80251]

According to pcpartpicker.com there's lots of 4 TiB SSD's but after that the capacities jump to 7.86 TiB and on up, there's nothing in-between. Why?

 

[Count von Schwalbe]

Doubling, I reckon. As capacities increase, increments decrease. Just slap two flash dies per one to upsize your drives.

 

[80251]

Count, did you really mean to write as capacities increase increments increase?

 

[80-watt Hamster]

This question has caused me to wonder about the economics/viability of various configurations. Let's say you've got 2.5" PCB with four pads for NAND chips. One could:

a) Populate all four with same-density chips, different densities for different capacities
b) Have one NAND module to rule them all, and fill varying numbers of pads
c) Combination of varying chip densities and pad placement

Based on currently available capacities, a) would suggest there being 32, 64, 256, 512 and 1024 GiB chips commonly in production. Option b) doesn't make much sense, actually, since that arrangement couldn't cover the capacities we see. So it's probably c). But is there anything preventing 3 of 4 pads occupied to get, say, a 640720 GB drive? Nobody's doing it, so controllers must not like chips that aren't arranged in powers of 2.

EDIT: Arithmetic'ed wrong: 3x 240 is 720, not 640

 

[80251]

@ 80-Watt Hamster
But there must be more than four NAND pads on SSD PCB's if the largest NAND chip size is 1024 GiB, because they have 7.86TiB and 8 TiB capacity SSD's and even larger.

They must be mixing chip densities if they manufacture 7.86TiB SSD's.

Personally, I'd like to see a 5 TiB or 6 TiB SSD because I don't foresee ever needing more storage space than that. The 4 TiB and up models are way too expensive right now AFAIC.

 

[P4-630]

Personally, I'd like to see a 5 TiB or 6 TiB SSD because I don't foresee ever needing more storage space than that.

And losing all data once de ssd gives up to work for some reason.....

 

[DoLlyBirD]

Count, did you really mean to write as capacities increase increments increase?

AS capacity (size) increases, the increments (space between the next sizes up) decreases, which makes perfect sense as you don't tend to see 768GB SSD, 512/1TB and so on and so forth, also SSD's are more related to RAM technology than HDD's and the same trend has been present with RAM for many years now: 512MB>1GB>2GB>4GB etc etc obviously there can be exceptions to the "rule" but that seems like the most plausible explanation, and the fact that RAM has always been manufactured in 1/2/4/8/16/32/64/128/256/512/1024 etc capactity

Doubling, I reckon. As capacities increase, increments decrease. Just slap two flash dies per one to upsize your drives.

@ 80-Watt Hamster
But there must be more than four NAND pads on SSD PCB's if the largest NAND chip size is 1024 GiB, because they have 7.86TiB and 8 TiB capacity SSD's and even larger.

They must be mixing chip densities if they manufacture 7.86TiB SSD's.

Personally, I'd like to see a 5 TiB or 6 TiB SSD because I don't foresee ever needing more storage space than that. The 4 TiB and up models are way too expensive right now AFAIC.

No, some MFR's are likely using 1000MB as a GB which will report in Windows less as Windows uses 1024MB as GB or vice versa, either way, your 8TB SSD doesn't show as 8TB in Windows

 

[80-watt Hamster]

AS capacity (size) increases, the increments (space between the next sizes up) decreases, which makes perfect sense as you don't tend to see 768GB SSD, 512/1TB and so on and so forth, also SSD's are more related to RAM technology than HDD's and the same trend has been present with RAM for many years now: 512MB>1GB>2GB>4GB etc etc obviously there can be exceptions to the "rule" but that seems like the most plausible explanation, and the fact that RAM has always been manufactured in 1/2/4/8/16/32/64/128/256/512/1024 etc capactity



No, some MFR's are likely using 1000MB as a GB which will report in Windows less as Windows uses 1024MB as GB or vice versa, either way, your 8TB SSD doesn't show as 8TB in Windows

Maybe I'm all wet, but aren't memory ICs forced to have a raw capacity of some power of two by their very nature? The usable capacity of the drive could be (and usually is) different.

@ 80-Watt Hamster
But there must be more than four NAND pads on SSD PCB's if the largest NAND chip size is 1024 GiB, because they have 7.86TiB and 8 TiB capacity SSD's and even larger.

They must be mixing chip densities if they manufacture 7.86TiB SSD's.

Personally, I'd like to see a 5 TiB or 6 TiB SSD because I don't foresee ever needing more storage space than that. The 4 TiB and up models are way too expensive right now AFAIC.

Four was just a hypothetical example. There could be eight pads, or 2TiB chips. I'm no expert.

 

[DoLlyBirD]

Maybe I'm all wet, but aren't memory ICs forced to have a raw capacity of some power of two by their very nature? The usable capacity of the drive could be (and usually is) different.



Four was just a hypothetical example. There could be eight pads, or 2TiB chips. I'm no expert.

You could be right, I remember something similar and tried to say that in my reply to the OP though, hot weather and alcohol/

 

[Count von Schwalbe]

Count, did you really mean to write as capacities increase increments increase?

The number decreases, the size increases.

This question has caused me to wonder about the economics/viability of various configurations. Let's say you've got 2.5" PCB with four pads for NAND chips. One could:

a) Populate all four with same-density chips, different densities for different capacities
b) Have one NAND module to rule them all, and fill varying numbers of pads
c) Combination of varying chip densities and pad placement

Based on currently available capacities, a) would suggest there being 32, 64, 256, 512 and 1024 GiB chips commonly in production. Option b) doesn't make much sense, actually, since that arrangement couldn't cover the capacities we see. So it's probably c). But is there anything preventing 3 of 4 pads occupied to get, say, a 640 GB drive? Nobody's doing it, so controllers must not like chips that aren't arranged in powers of 2.

Flash controllers have a certain number of NAND channels available, I would think it would decrease performance to not populate all of those channels. Also, there may be firmware issues. Most likely there are dies with a certain interface of varying densities.

Also, if you look at the models with 7.86 TB capacities, say the Kingston DC450R, it comes in 480GB to 7680GB capacities, aka 240GB*2^x.

Ok, so looking at it further, the DC450R is using an eight-channel controller, and has eight packages of flash onboard. All models sustain similar read/write speeds, but being SATA may affect this. The dies in the packages are 512 Gbit 3D NAND from Kioxia. Assuming 4 dice per package, we get 2048 GB. As the available capacity is 1920 GB, and this is an enterprise drive, that means 7.25% of the capacity is hidden for overprovisioning.

This means that this drive comes with packages (stacks) of flash with 1, 2, 4, 8, and 16 dies.

 

[joemama]

This could also be a market issue, most consumers/computer vendors only purchase volumes of 128GB to 2TB depending on their budget, thus it is worth to split finer for this volume area.
Over 2TB I would guess the market is mainly enterprise oriented rather than normal consumers and the budget range is a lot wider so there is no benefit of splitting into too many different volumes.