Sunday, November 9th 2025

Samsung Teases Modular SSD and M.2 2242 SSD with DRAM

Although we're almost two months away from CES, companies are already starting to tease products that they will have on display at the show. Samsung has unveiled a pair of new SSDs, of which one might find an appeal with owners of portable gaming devices. The other, is not going to be something most consumers will ever have any direct contact with, but the Detachable AutoSSD is an interesting product from Samsung that might spill over in other markets in the future. It's intended for the automotive market and uses the E1.A form factor, but what makes it truly unique is the fact that the SSD controller and the NAND are separated from each other, something Samsung claims improves heat dissipation, but also that it can extend the product life and make upgrades and/or replacements easier.

A more useful addition to its lineup of SSDs is the PM9E1 M.2 22x42, which is an OEM drive in the M.2 2242 form factor. This is as far as we're aware, the first announcement of an M.2 2242 PCIe 5.0 drive, but it's also one of what has to be less than a handful of drives in this form factor that has DRAM onboard. Samsung claims read speeds of up to 14.8 GB/s and write speeds of up to 13.4 GB/s, courtesy of its Presto SSD controller and its V8 TLC V-NAND. The PM9E1 will come in sizes up to 4 TB, which means it's a double sided drive with components on both sides of the PCB, which could make it a tad tricky when it comes to the cooling. However, for those with a handheld gaming device like the Steam Deck, could prove to be one of the best performance options out there, while at the same time offering plenty of storage space for your game library.
Source: Samsung Semiconductors
Add your own comment

21 Comments on Samsung Teases Modular SSD and M.2 2242 SSD with DRAM

#1
shuffy_losys
The question is...

Do gaming machines even need 14.8GB/s reads yet? I still haven't found a game that doesn't take more than 10 or so seconds to load a save file from a decent SSD and I'm pretty sure most of that time is the CPU chugging at getting everything setup.
Posted on Reply
#2
Raiden85
14.8GB/s is way more than enough speed for many years to come. I just wish they would work on providing significantly larger capacities that don't cost the price of a car.

SSD capacity on the consumer side has stagnated for way too long. You have Kioxia with a 245TB 2.5' drive on the enterprise side that can still do an overkill read of 12GB/s, while we're stuck with 8TB that still costs way too much. Where’s my 16TB, 32TB, or 64TB NVMe at decent prices? Consumers are being mugged.

I bought a Samsung 8TB QVO drive a few years ago, and it's still the largest consumer capacity which is bloody insane.
Posted on Reply
#3
shuffy_losys
Raiden85SSD capacity on the consumer side has stagnated for way too long. You have Kioxia with a 245TB 2.5' drive on the enterprise side that can still do an overkill read of 12GB/s, while we're stuck with 8TB that still costs way too much. Where’s my 16TB, 32TB, or 64TB NVMe at decent prices? Consumers are being mugged.
It's also at roughly the same $0.65 per gigabyte pricing (compared to say a Samsung 990 EVO) (EDIT: pricing info I found was speculative)

The problem with flash memory is reliability issues at smaller process nodes, hence why they've resorted to 3D stacking. So until we ever overcome this or find another way to store data, we're pretty much stuck here.
Posted on Reply
#4
FreedomEclipse
~Technological Technocrat~
shuffy_losysThe question is...

Do gaming machines even need 14.8GB/s reads yet? I still haven't found a game that doesn't take more than 10 or so seconds to load a save file from a decent SSD and I'm pretty sure most of that time is the CPU chugging at getting everything setup.
I think its currently limited to specific scenarios. i.e a skyrim/Elder scrolls/Fallout game with bajillion 4k texture mods and extra crispy community made content etc etc.
Posted on Reply
#5
RejZoR
shuffy_losysThe question is...

Do gaming machines even need 14.8GB/s reads yet? I still haven't found a game that doesn't take more than 10 or so seconds to load a save file from a decent SSD and I'm pretty sure most of that time is the CPU chugging at getting everything setup.
Reality is, despite all the hype around DirectStorage and how games will now be able to stream multiple gigabytes of data in seconds, reality is, if you just have a decent SATA SSD and I don't think anyone will be able to tell a difference. HDD's were horrible because randomly accessing texture data bits from a 50GB texture file quickly turned into seconds long access times paired with sub 50MB/s reads because of it. SATA SSD has 550MB/s sequential and its slight IOPS disadvantage compared to NVMe doesn't really show up with games.

Just having ANY kind of SSD that at least has onboard DRAM and you'll be fine. And I don't expect this to change any time soon. I have 2TB SATA SSD and 4TB NVMe SSD (both top end in its class). For testing not long ago I moved everything to SATA SSD. I couldn't really tell a difference. If you have NVMe PCIe 3.0 drive you're more than fine today and you probably have a faster drive than 75% of computer users.
Posted on Reply
#6
chrcoluk
Games have to be coded to take advantage of NVME, which is why 99% of them perform pretty much the same SATA SSD vs NVME SSD.

Nier replicant is an interesting one, from a JRPG dev, JRPG devs have a habit of carrying on coding the game after release, and then "not" back porting the improvements to the earliest versions.

Nier Replicant was released on PS4 first and also plays on PS5 the PS4 build, it has no AF, bilinear instead, and wait for it longer loading times. This version feels like a straight PS4 compat port on PS5.
On both Xbox Series and PC which came later they got AF, and much better loading times. Optimised for the platforms.

I dont think its directstorage level of optimisation, but clearly something was done on the loading process, as e.g. the home village is 2 seconds on PC/Xbox and 9 seconds on PS5.

I think even if we think about media editing, directstorage games (as rare as they are at the moment), its hard to consider gen 5 as an exciting development, I feel what most people care about at this point is cost and capacity. Gen 4 is easily good enough,

We can definitely get double digit capacity in 2.5inch SSD's that issue is a combination of high margin expectation and protecting the enterprise market. I expect it should be possible to do 8TB 2280 as well.
Posted on Reply
#7
shuffy_losys
chrcolukGames have to be coded to take advantage of NVME, which is why 99% of them perform pretty much the same SATA SSD vs NVME SSD.
By "take advantage", the only thing they'd have to do is slurp gigabytes per second of data constantly, which isn't happening. It's storage, there's nothing programmatically special you need to do other than consume more data per unit of time to take advantage of faster storage performance.

A lot of game engines also mitigate loading hitches by loading the bare minimum first, then load in the higher quality assets later on spare cycles.
Posted on Reply
#8
Wirko
shuffy_losysBy "take advantage", the only thing they'd have to do is slurp gigabytes per second of data constantly, which isn't happening. It's storage, there's nothing programmatically special you need to do other than consume more data per unit of time to take advantage of faster storage performance.
Indeed, I don't understand what's so special about NVMe. But games and everything else alike would have to make better use of queued reads. There's a lot of SSD performance left unused yet.
Posted on Reply
#9
londiste
WirkoIndeed, I don't understand what's so special about NVMe.
Lower overhead, both in terms of drive being closer to CPU as well as in terms of protocol?
Posted on Reply
#10
Wirko
londisteLower overhead, both in terms of drive being closer to CPU as well as in terms of protocol?
Lower overhead can't improve the weakest point of all NAND-based SSDs, which is the random read speed of about 32k IOPS (improved from ~25k IOPS in the last couple years). Raise the queue depth to 32 and you get 32x the speed - with the same overhead, communication latency and transfer speed.
Posted on Reply
#11
Wales
shuffy_losysThe question is...

Do gaming machines even need 14.8GB/s reads yet? I still haven't found a game that doesn't take more than 10 or so seconds to load a save file from a decent SSD and I'm pretty sure most of that time is the CPU chugging at getting everything setup.
The read/write speeds you see are merely the performance limits of the SSD controller, not the actual performance of the underlying NAND flash.
In reality, if you remove the SLC cache trick, the true write performance of TLC flash can’t even exceed 500 MB/s.
Posted on Reply
#12
Chaitanya
This should be a decent upgrade over existing 2242 sized drives.
Posted on Reply
#13
chrcoluk
shuffy_losysBy "take advantage", the only thing they'd have to do is slurp gigabytes per second of data constantly, which isn't happening. It's storage, there's nothing programmatically special you need to do other than consume more data per unit of time to take advantage of faster storage performance.

A lot of game engines also mitigate loading hitches by loading the bare minimum first, then load in the higher quality assets later on spare cycles.
DS is an api, i/o has to specifically call it, also taking advantage of extra threads. Nier replicant I dont think is even using DS, very likely isnt, what they probably did in that game is make it multi threaded for the loading on PC/Xbox, but same work wasnt done on PS5 so PS5 is only little faster than PS4.
Posted on Reply
#14
londiste
WirkoLower overhead can't improve the weakest point of all NAND-based SSDs, which is the random read speed of about 32k IOPS (improved from ~25k IOPS in the last couple years). Raise the queue depth to 32 and you get 32x the speed - with the same overhead, communication latency and transfer speed.
Fair enough but lower overhead is what is special in NVMe :)
And big sequential speeds have very little actual use but are good form marketing and come with relatively small effort on state of the art drives.

Isn't random performance largely down to cost and efficiency though? Flash does have its limitations but more channels and more Flash chips plus more RAM for caching has effectively been the answer to this but the cost-performance benefit for sales isn't there. I do wish XPoint/Optane would still get R&D but that does seem to be gone and alternatives all have an uphill battle against mass-produced Flash that is cheap thanks to economies of scale.
Posted on Reply
#15
EsaT
shuffy_losysThe problem with flash memory is reliability issues at smaller process nodes, hence why they've resorted to 3D stacking. So until we ever overcome this or find another way to store data, we're pretty much stuck here.
"3D" NAND precisely took node size back multiple steps, basically making TLC equal to old planar MLC.
Since their first appearance amount of layers has increased steadily increasing capacity per chip.
Basically we're now around 300 layer level and manufacturers are racing toward 400 layers:
www.allpcb.com/allelectrohub/3d-nand-hits-400-layers-with-hybrid-bonding
WalesThe read/write speeds you see are merely the performance limits of the SSD controller, not the actual performance of the underlying NAND flash.
In reality, if you remove the SLC cache trick, the true write performance of TLC flash can’t even exceed 500 MB/s.
TLC drives have long ago exceeded 500 MB/s write speeds.
Full drive fill test in TPU's reviews tests precisely that by saturating drive so that it doesn't have spare time to empty SLC cache, but has to go writing natively in TLC.

www.techpowerup.com/review/team-group-mp34-nvme-ssd/6.html
www.techpowerup.com/review/western-digital-wd-black-2018-nvme-500-gb/6.html

It's QLC drives which struggle to exceed HDDs in native write speed.
Posted on Reply
#16
Patriot
EsaT"3D" NAND precisely took node size back multiple steps, basically making TLC equal to old planar MLC.
Since their first appearance amount of layers has increased steadily increasing capacity per chip.
Basically we're now around 300 layer level and manufacturers are racing toward 400 layers:
www.allpcb.com/allelectrohub/3d-nand-hits-400-layers-with-hybrid-bonding


TLC drives have long ago exceeded 500 MB/s write speeds.
Full drive fill test in TPU's reviews tests precisely that by saturating drive so that it doesn't have spare time to empty SLC cache, but has to go writing natively in TLC.

www.techpowerup.com/review/team-group-mp34-nvme-ssd/6.html
www.techpowerup.com/review/western-digital-wd-black-2018-nvme-500-gb/6.html

It's QLC drives which struggle to exceed HDDs in native write speed.
Look at the review you linked, and look at Q-depth 1 performance.
That is the performance writing to 1 stack of nand... 50k iops of 4kb.
That maths out to.... 200MB/s
Drives have many channels and you have to write to multiple to get good performance because the underlying nand has not scaled.
Posted on Reply
#17
remixedcat
wonder what the temperature and moisture tolerances would be on that samsung drive... would be cool in a rugged pc.
Posted on Reply
#18
Wirko
PatriotLook at the review you linked, and look at Q-depth 1 performance.
That is the performance writing to 1 stack of nand... 50k iops of 4kb.
That maths out to.... 200MB/s
Drives have many channels and you have to write to multiple to get good performance because the underlying nand has not scaled.
No, this is the performance writing to SLC cache. We don't know but I suspect that the SSD controllers also have a small amount of static RAM for various purposes, including some write caching.
- Real TLC writing is roughly 2k IOPS (1 divided by NAND program time of ~500 µs, which you can find for some NAND types in the TPU SSD DB)
- writing with fsync is up to 8k IOPS, no parallelisation is possible here (unless the controller is cheating), it can't be to TLC, therefore it's to SLC - see the latest SSD review (Phison E28)
- Random Q1 writing is above 100k IOPS, also from the latest review; this may involve SLC writing, a lot of parallelisation, reordering of writes for best performance and who knows what else.

One of my recent comments is actually closely related to this:
Wirko in Phison E28 PCIe 5.0 SSD Engineering SampleMaybe it's not lying. Writing to SLC cache could actually be that fast. As an example, the WD SN850X (2TB) with BiCS5 NAND chips has a program time of 484 µs, and I understand that this means TLC. SLC…
Posted on Reply
#19
remixedcat
chrcolukGames have to be coded to take advantage of NVME, which is why 99% of them perform pretty much the same SATA SSD vs NVME SSD.
what do you have to do to code a game to take advantage of nvme?
Posted on Reply
#20
maxli86
I really hope Samsung should re-release SATA and PCIE Gen 3 SSD.
I did write to them about this as do any application really need that fast.
Even the Samsung 990 Pro for most users you won't tap on it's 7900MB/S
unless you are doing editing and heavy transfer.

I did enquire will there be anymore Gen 3 SSD especially for older PCs out there.
Not many will know Gen 3 drives Samsung 970 EVO Plus have been discontinued and
soon Samsung 980 (Non-pro).

These drives even though the read/write is at 3500mb/s so actually more than
sufficient for most users.
And yes Gen 4 is backward compatible but kind of waste in my opinion.

M.2 SATA are non-existent except PCs that have such limitation.
So Samsung not everyone need really fast SSD unless AI or Editing purpose.
Some just want fast and reliable SSD for data and games.
Posted on Reply
#21
chrcoluk
remixedcatwhat do you have to do to code a game to take advantage of nvme?
chrcoluk in Samsung Teases Modular SSD and M.2 2242 SSD with DRAMDS is an api, i/o has to specifically call it, also taking advantage of extra threads. Nier replicant I dont think is even using DS, very likely isnt, what they probably did in that game is make it mu…
Similar to things like IO_DIRECT.

This is common on i/o, there is different ways of making requests.
Posted on Reply
Add your own comment
Nov 15th, 2026 05:12 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts