Yes but one would hope that if there are working prototypes, then by now we'd see them somewhere? That doesn't seem to be the case, not even rumors IIRC.
Fair enough. I'm not sure what is going on. Maybe they're trying to play things differently from Samsung and are going to hide their Optane competitor until it is ready to go?
More likely than not they are looking at how much of a loosing proposition that ReRam, XPointe, and other such 'between RAM and storage' solutions are, and are just sitting on it. Keep it on the calendar to make them look competitive with Intel to shareholders, but never make a move to prevent them from loosing money on a niche technology that very few people are actually asking for.
3DXpoint is actually a very valuable technology. If Intel can lower the cost for the second generation, it would become far less of a niche product, and it already has a fairly sizable market right now.
Halving the price would begin to let it cut into the SSD market directly; especially if they can get the power low enough to make it a reasonable option for laptops. A $110 for 120GB 3x point module would be big enough to serve as the only drive on a lot of mainstream systems; and is affordable enough to make dual partition enthusiast setups viable; similar to how many of us had a 120/240GB SSD and an HDD a few years back when 512GB of flash was prohibitively expensive.
A second halving of price would put it at the same ~1TB for $500 price range when I went pure SSD for my main system, and would be borderline viable for my switching over it when I build a new system in a few years. (Depends on if software bloat has me thinking I'd need 2TB to be safe or not.)
A $110 for 120GB 3x point module would be big enough to serve as the only drive on a lot of mainstream systems; and is affordable enough to make dual partition enthusiast setups viable;
Sorry but with the advent of QLC NAND, it's rated endurance (about ~1k best case) & the ensuing price war (China is betting big on NAND & DRAM) there's no way 3dXP can do all of that at an affordable price point. The scenario you're describing is a real oddball, basically you're thinking of someone who needs better speeds (also endurance?) than the best NVMe drives out there, yet doesn't need more than 120GB (or more) & is willing to pay a premium for a *slight increase in speeds. When you have flagship phones selling with 512GB storage (rumored with Note9) there's no way a mainstream laptop will look good with just 120GB on board.
*depending on usage the difference might be substantial, but still needs lots of compromise especially wrt storage space.
the would be nice to, and if they can/able/willing to are very vastly different things.
As far as I understood Xpoint compared to "standard" nand flash is quite expensive to produce, so likely they only have so much "wiggle room" are end up selling it "at cost" which the shareholders etc would be PISSED if they did this.
(kind of like HBM because of the interposer ends up being more expensive because there are extra steps required, more certification, more precision vs just printing off a "bare chip" that either makes the spec required or does not, which means can become flash for a USB stick, SSD, full blown NVME instead...I think IMO Xpoint likely is very similar ball game, is not widespread use (cannot be used everywhere) same as HBM cannot, so therefore higher than normal price along with more complexity behind the design as well)
If anything makers of NVME style likely are raking in handsome profits (on overall amounts sold) because they are "fast" not because they are likely any more expensive in comparison to the "old" style SSD...
they have went SLC-MLC-TLC-3d-3d Vnand-QLC (soon) etc, so the controllers are likely dirt cheap compared to what they once were AND they save quite a bit by not having to worry about heatsinks etc, just a bare pcb with a couple of chips on it (with u2/m2 vs standard 2.5mm SSD sata based)
3D XPoint is aptly named, i.e. no point for most users currently. I think it is a great SSD/RAM hybrid, but it is more SSD than RAM. The other unfortunate points are the price and the power draw. At low capacity, the power draw is pretty high, and the higher capacity ones draws too much power and is reaching the point where it needs a chunky if not active heatsink.
When it ends up as a 4GB cache for a 2TB QLC PCEex4 SSD, and they update the controller to take full advantage of the PCIex4, nobody would complain about 4GB/s speeds on most used files on a cheap (per TB) SSD.
QLC is not a thrilling idea to me. The additional capacity is nice, but I was hoping we'd have a more stable/durable replacement for NAND before QLC approached mainstream sales. It really isn't even endurance that worries me at this point. It's data retention over time that makes me hesitate. I want to go pure solid state because I really like the performance benefits and resistance to shock and impact, but I don't feel like I should rely on SSDs for long term, powered-down storage. I've already dealt with data corruption using 2.5 inch SATA SSDs in external casings and have since switched back to platter-based drives for backups as a result. QLC seems perhaps more susceptible to loss than existing NAND, but I can't have a butterfingers moment and drop my backup HDD on the floor so I almost panic when I carry that external disk around.
"Quality" Nand does not appear to have a durability issue, only a data retention issue
I have been running Killdisk on a cheap 8GB Silicon Power thumb drive for almost 4 years now and it still works fine
I should reach my goal of 350 full wipes in the next few days
Killdisk has changed its website since I began testing the durability of Nand and they now proudly claim that Killdisk "IS" compatible with Flash memory
likely Samsung/TSMC will hit 3nm VERY soon, as far as "layers" are concerned, am sure the nm "shrinks" will hit a cannot pass plateau very soon (until beyond silicon is able to be done at a low cost such as pure optical to optical) so they may have to resort to "happy with 7nm or 5 or 3nm and instead add layer upon layer to increase durability/capacity and in a "round about" way endurance (data retention, P/E cycles)
suppose in reality it was not that long ago we were at 150+nm, then 45nm, then 32-20-14-10 and very very soon 7nm, so it seems because of the massive $$$$$$$$$$$$$$ they are making on these things, the transitions to new tech is happening VERY rapidly indeed.
now 2048 layer.....HAHAHA, I do not picture this anytime soon, likely before they come to this point "layers" are likely to have been "ancient" technology done in the 2010s...I can picture in the ~2025 likely will be fairly close to being "optical" throughout (many companies have been working/refining this for the past decade or more)
They can "hit" it today, after all, these are just marketing names and have nothing to do with reality. But then they would need to invent and market a new naming scheme in a year or two, and their current "10nm" and "7nm" will be harder to sell.
NAND went 3D because it hit the limit on horizontal scaling. 2040 layers in 10-15 years if NAND survives for that long so no cheaper solution comes up. There will likely be more than 4 bits per cell too by then.
lol see what happens if you aren't paying attention
This was said on January 25 " Additionally, we commenced initial production of our 96-layer technology, BiCS4, and began product shipments to retailers in the December quarter. We expect to ramp BiCS4 in the second half of the calendar year. "
I think it is a meaningless race to produce a higher capacity SSD with significant deterioration of endurance. Understandably that even current TLC NAND, especially the 3D type in theory could last a very long time for most normal usage, but this leads to the next question, in terms of cost savings, I also question if QLC will be a significant savings over TLC SSDs.
They're hitting 1k P/E cycles with QLC, that's as good as a lot of first gen TLC flash; and outside of very write intensive work loads (which shouldn't be using cheap mass market SSDs anyway) will last long enough that something else failing will kill the drive before it hits its max wear level.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
29 Comments
Back to Article
R0H1T - Tuesday, May 29, 2018 - link
I'm guessing no one asked them about the vaporware *ReRAM?*in development since temps immémorial.
MajGenRelativity - Tuesday, May 29, 2018 - link
It's not supposed to land until the end of the year according to that chartR0H1T - Tuesday, May 29, 2018 - link
Yes but one would hope that if there are working prototypes, then by now we'd see them somewhere? That doesn't seem to be the case, not even rumors IIRC.MajGenRelativity - Tuesday, May 29, 2018 - link
Fair enough. I'm not sure what is going on. Maybe they're trying to play things differently from Samsung and are going to hide their Optane competitor until it is ready to go?CaedenV - Tuesday, May 29, 2018 - link
More likely than not they are looking at how much of a loosing proposition that ReRam, XPointe, and other such 'between RAM and storage' solutions are, and are just sitting on it. Keep it on the calendar to make them look competitive with Intel to shareholders, but never make a move to prevent them from loosing money on a niche technology that very few people are actually asking for.MajGenRelativity - Tuesday, May 29, 2018 - link
3DXpoint is actually a very valuable technology. If Intel can lower the cost for the second generation, it would become far less of a niche product, and it already has a fairly sizable market right now.ZeDestructor - Tuesday, May 29, 2018 - link
The other part of that is that we're still waiting on Intel/Micron to ship Optane-based NVDIMMswanderer66 - Wednesday, May 30, 2018 - link
Almost certain that's a power-draw issue at this point.DanNeely - Tuesday, May 29, 2018 - link
Halving the price would begin to let it cut into the SSD market directly; especially if they can get the power low enough to make it a reasonable option for laptops. A $110 for 120GB 3x point module would be big enough to serve as the only drive on a lot of mainstream systems; and is affordable enough to make dual partition enthusiast setups viable; similar to how many of us had a 120/240GB SSD and an HDD a few years back when 512GB of flash was prohibitively expensive.A second halving of price would put it at the same ~1TB for $500 price range when I went pure SSD for my main system, and would be borderline viable for my switching over it when I build a new system in a few years. (Depends on if software bloat has me thinking I'd need 2TB to be safe or not.)
MajGenRelativity - Tuesday, May 29, 2018 - link
Halving the price would definitely be a big step, especially with a power consumption decrease. Hopefully Intel can make some steps in that direction.R0H1T - Tuesday, May 29, 2018 - link
A $110 for 120GB 3x point module would be big enough to serve as the only drive on a lot of mainstream systems; and is affordable enough to make dual partition enthusiast setups viable;Sorry but with the advent of QLC NAND, it's rated endurance (about ~1k best case) & the ensuing price war (China is betting big on NAND & DRAM) there's no way 3dXP can do all of that at an affordable price point. The scenario you're describing is a real oddball, basically you're thinking of someone who needs better speeds (also endurance?) than the best NVMe drives out there, yet doesn't need more than 120GB (or more) & is willing to pay a premium for a *slight increase in speeds. When you have flagship phones selling with 512GB storage (rumored with Note9) there's no way a mainstream laptop will look good with just 120GB on board.
*depending on usage the difference might be substantial, but still needs lots of compromise especially wrt storage space.
smilingcrow - Tuesday, May 29, 2018 - link
Q: What has the storage capacity of a phone, which is primarily for media files got to do with the capacity of a boot drive for Windows?A: Nothing.
R0H1T - Tuesday, May 29, 2018 - link
So you're gonna sell 120GB 3dXP on a mainstream laptop when 256~512GB SSD will be much cheaper, even for OEMs?Newsflash 256|512>>>120 especially for the masses!
surt - Tuesday, May 29, 2018 - link
You have to remember to show the masses your bar graph that shows your laptop is twice as fast** as the competition.** In contrived benchmark designed to show 3dx in its best light.
Dragonstongue - Tuesday, May 29, 2018 - link
the would be nice to, and if they can/able/willing to are very vastly different things.As far as I understood Xpoint compared to "standard" nand flash is quite expensive to produce, so likely they only have so much "wiggle room" are end up selling it "at cost" which the shareholders etc would be PISSED if they did this.
(kind of like HBM because of the interposer ends up being more expensive because there are extra steps required, more certification, more precision vs just printing off a "bare chip" that either makes the spec required or does not, which means can become flash for a USB stick, SSD, full blown NVME instead...I think IMO Xpoint likely is very similar ball game, is not widespread use (cannot be used everywhere) same as HBM cannot, so therefore higher than normal price along with more complexity behind the design as well)
If anything makers of NVME style likely are raking in handsome profits (on overall amounts sold) because they are "fast" not because they are likely any more expensive in comparison to the "old" style SSD...
they have went SLC-MLC-TLC-3d-3d Vnand-QLC (soon) etc, so the controllers are likely dirt cheap compared to what they once were AND they save quite a bit by not having to worry about heatsinks etc, just a bare pcb with a couple of chips on it (with u2/m2 vs standard 2.5mm SSD sata based)
watzupken - Wednesday, May 30, 2018 - link
3D XPoint is aptly named, i.e. no point for most users currently. I think it is a great SSD/RAM hybrid, but it is more SSD than RAM. The other unfortunate points are the price and the power draw. At low capacity, the power draw is pretty high, and the higher capacity ones draws too much power and is reaching the point where it needs a chunky if not active heatsink.peevee - Wednesday, May 30, 2018 - link
When it ends up as a 4GB cache for a 2TB QLC PCEex4 SSD, and they update the controller to take full advantage of the PCIex4, nobody would complain about 4GB/s speeds on most used files on a cheap (per TB) SSD.PeachNCream - Tuesday, May 29, 2018 - link
QLC is not a thrilling idea to me. The additional capacity is nice, but I was hoping we'd have a more stable/durable replacement for NAND before QLC approached mainstream sales. It really isn't even endurance that worries me at this point. It's data retention over time that makes me hesitate. I want to go pure solid state because I really like the performance benefits and resistance to shock and impact, but I don't feel like I should rely on SSDs for long term, powered-down storage. I've already dealt with data corruption using 2.5 inch SATA SSDs in external casings and have since switched back to platter-based drives for backups as a result. QLC seems perhaps more susceptible to loss than existing NAND, but I can't have a butterfingers moment and drop my backup HDD on the floor so I almost panic when I carry that external disk around.dgingeri - Tuesday, May 29, 2018 - link
I totally agree. I prefer quality over quantity. Hopefully, we'll get a technology soon that doesn't have the durability issues NAND has.ಬುಲ್ವಿಂಕಲ್ ಜೆ ಮೂಸ್ - Tuesday, May 29, 2018 - link
I must disagree dgingeri -"Quality" Nand does not appear to have a durability issue, only a data retention issue
I have been running Killdisk on a cheap 8GB Silicon Power thumb drive for almost 4 years now and it still works fine
I should reach my goal of 350 full wipes in the next few days
Killdisk has changed its website since I began testing the durability of Nand and they now proudly claim that Killdisk "IS" compatible with Flash memory
Durability does not seem to be the issue!
dgingeri - Tuesday, May 29, 2018 - link
So, when are we getting the 3nm 2048 layer QLC 3D NAND?R0H1T - Tuesday, May 29, 2018 - link
So, when are we getting the 3nm 2048 layer QLC 3D NAND?2048 presumably!
Dragonstongue - Tuesday, May 29, 2018 - link
likely Samsung/TSMC will hit 3nm VERY soon, as far as "layers" are concerned, am sure the nm "shrinks" will hit a cannot pass plateau very soon (until beyond silicon is able to be done at a low cost such as pure optical to optical) so they may have to resort to "happy with 7nm or 5 or 3nm and instead add layer upon layer to increase durability/capacity and in a "round about" way endurance (data retention, P/E cycles)suppose in reality it was not that long ago we were at 150+nm, then 45nm, then 32-20-14-10 and very very soon 7nm, so it seems because of the massive $$$$$$$$$$$$$$ they are making on these things, the transitions to new tech is happening VERY rapidly indeed.
now 2048 layer.....HAHAHA, I do not picture this anytime soon, likely before they come to this point "layers" are likely to have been "ancient" technology done in the 2010s...I can picture in the ~2025 likely will be fairly close to being "optical" throughout (many companies have been working/refining this for the past decade or more)
peevee - Wednesday, May 30, 2018 - link
"likely Samsung/TSMC will hit 3nm VERY soon"They can "hit" it today, after all, these are just marketing names and have nothing to do with reality. But then they would need to invent and market a new naming scheme in a year or two, and their current "10nm" and "7nm" will be harder to sell.
jjj - Tuesday, May 29, 2018 - link
NAND went 3D because it hit the limit on horizontal scaling.2040 layers in 10-15 years if NAND survives for that long so no cheaper solution comes up.
There will likely be more than 4 bits per cell too by then.
eddieobscurant - Tuesday, May 29, 2018 - link
Is toshiba's/wd's 96-layer string stacking like intel's/micron's or a classic one like samsung's ?jjj - Tuesday, May 29, 2018 - link
lol see what happens if you aren't paying attentionThis was said on January 25 " Additionally, we commenced initial production of our 96-layer technology, BiCS4, and began product shipments to retailers in the December quarter. We expect to ramp BiCS4 in the second half of the calendar year. "
watzupken - Wednesday, May 30, 2018 - link
I think it is a meaningless race to produce a higher capacity SSD with significant deterioration of endurance. Understandably that even current TLC NAND, especially the 3D type in theory could last a very long time for most normal usage, but this leads to the next question, in terms of cost savings, I also question if QLC will be a significant savings over TLC SSDs.DanNeely - Wednesday, May 30, 2018 - link
They're hitting 1k P/E cycles with QLC, that's as good as a lot of first gen TLC flash; and outside of very write intensive work loads (which shouldn't be using cheap mass market SSDs anyway) will last long enough that something else failing will kill the drive before it hits its max wear level.