RejZoRYes it does. You can't crank up MLC prices because your QLC process is shit. If QLC devices are more expensive because of it, whatever. But if you jack up prices of MLC or TLC because of it, then you can kindly fuck off as a vendor. Especially if that becomes a thing just globally across entire market for all vendors because one can't make them properly.

TheLostSwedeSo if it's a bad corn harvest, farmers should charge more for wheat? I don't think that's how the real world works...

I was kinda referring to QLC prices, but you have a point.

RejZoRThey may, but then no one will buy it. It's not much different with NAND. But they want to apply the logic of "corn harvest was bad, so we need to jack up the prices of beans and potatoes as well to level things out for us". No wonder ppl hate them lol

They'll buy it .. if everyone raises the prices. ;)

nemesis.ieBad analogy, both my bike run very nicely on 2 cylinders thanks (they only have two (each). :)

Sure but those only carry a single person ;) Not a whole marketplace.

not a surprise considering new tech always have low yields. Look at AMD for a good comparison... they too struggled in diffusing their Vega cores with HBM v2 with both V56 & V64.

the need...
the greed.....
and now, the deed........

all is well in the world of capitalism 101 :D

nOt

Vayra86Sure but those only carry a single person ;) Not a whole marketplace.

Well they can carry 4 people between them, but I get your point.

@Tsukiyomi91 Your Vega comparison is good, except that AMD did not jack up the prices as a result, with some even claiming they sold them at a loss.

''a Prelude to Global SSD Price Hikes?''

should of been stated as ''a Prelude to Global SSD Price Fixing? ''

now all they need is someone to start a factory production fire and the fix is in...

bonehead123the need...
the greed.....
and now, the deed........

all is well in the world of capitalism 101 :D

nOt

If people buy it at higher prices then all is indeed well.

nemesis.ieWell they can carry 4 people between them, but I get your point.

@Tsukiyomi91 Your Vega comparison is good, except that AMD did not jack up the prices as a result, with some even claiming they sold them at a loss.

Lucky they sold them at all, lol. I almost bought one, too expensive for my needs.

Tsukiyomi91not a surprise considering new tech always have low yields. Look at AMD for a good comparison... they too struggled in diffusing their Vega cores with HBM v2 with both V56 & V64.

This, its a new product, I dont think it will affect prices much if at all. Products may be waiting to be used with it, but its not liek they forgot or shut down all the other NAND assembly lines.

The whole nand industry is based on price fixing.

btarunrThe first victim of low yields of 3D QLC NAND flash is Intel's SSD 660p series, a mainstream NVMe SSD that brought 1 TB of storage under the $200-mark.

The thing i don't understand here is that as mentioned Intel's SSD 660p is some of the first ( if not the first ) NVMe SSD to bring 1 TB of storage for under $200 wich means that yields where good enough to allow this to happen right ? Considering that yields can only improve with time then i don't get how we end up with this situation ..... makes 0 sense ! This smell fishy to say the least .

so somebody screwed up at the factory, low yields means production errors, errors made by humans, and now we, the consumer, will pay for those mistakes. Not exactly something we can fight back against.

Isn't SLC the most reliable? I thought the more levels meant higher capacity at a cost of reliability?

@nemesis.ie not to mention that Vega is a little too late into the game when Pascal obliterated them in the high end market segment; no offense.
Anyhow, since we're seeing new techs, we'll have to give them time to mature before we can actually see how much value we're getting out of them over what we currently have.

Let's see if Sammy, SK Hynix, Toshiba, Sandisk etc. follow suits and declare poor yield as well. If not, then it's just an isolated case because of incompetence of Intel/IMFT. Otherwise we can expect an incoming price fix on SSDs.

And what's QLC cell average read & write before dying ? 500 ? 700 ? so it's a 4 year lifespan SSD ?
TLC is already 1000 to 2000 rewrites before dying, what is QLC made for ?

Sauce : www.anandtech.com/show/6459/samsung-ssd-840-testing-the-endurance-of-tlc-nand
(old, but still)

Genuine Q: Why is everyone using "Sauce" instead of "Source" when quoting sources these days? Is it some kind of "in joke" I missed?

nemesis.ieGenuine Q: Why is everyone using "Sauce" instead of "Source" when quoting sources these days? Is it some kind of "in joke" I missed?

more than likely auto-correction from a phone.

RenaldAnd what's QLC cell average read & write before dying ? 500 ? 700 ? so it's a 4 year lifespan SSD ?
TLC is already 1000 to 2000 rewrites before dying, what is QLC made for ?

Sauce : www.anandtech.com/show/6459/samsung-ssd-840-testing-the-endurance-of-tlc-nand
(old, but still)

Yes it's old & not valid for 3d NAND, IIRC 850 EVO was rated at 3k p/e cycles? The best estimates for QLC put it @1k atm which could change in the future, as we see more endurance tests like this one ~ techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

@DeathtoGnomes It looks BTA uses his phone for entering/editing some of his content so. ;)

R0H1TYes it's old & not valid for 3d NAND, IIRC 850 EVO was rated at 3k p/e cycles? The best estimates for QLC put it @1k atm which could change in the future, as we see more endurance tests like this one ~ techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

Thx, found couple article on Hardware.fr back in the days and ZDNET.
All are stating between 500 and 1000 p/e cycles, even by Toshiba themselves ; not higher. Pretty sure endurance is higher now, at same shrink node, but, they are going onto 10 and 7nm (even though it's not real 7nm, but that's not the point).
TLC even with 3D-NAND gets less and less reliable every time they shrink. And that also why Intel is strugelling right now.

I'm quite afraid of an 256GB SSD with QLC...1TB you can see it coming, but 256GB or less ...

RenaldThx, found couple article on Hardware.fr back in the days and ZDNET.
All are stating between 500 and 1000 p/e cycles, even by Toshiba themselves ; not higher. Pretty sure endurance is higher now, at same shrink node, but, they are going onto 10 and 7nm (even though it's not real 7nm, but that's not the point).
TLC even with 3D-NAND gets less and less reliable every time they shrink. And that also why Intel is strugelling right now.

I'm quite afraid of an 256GB SSD with QLC...1TB you can see it coming, but 256GB or less ...

Where are you hearing that QLC is going to made on the 10/7nm process node?

Yield is different from reliability. They probe wafers, test them, and say ok, this wafer passed, this wafer failed (at a high level).

Reliability is can the chip work long long term. Your chip can yield, but it might not be reliable.

The article says that they're having yield problems.

RenaldThx, found couple article on Hardware.fr back in the days and ZDNET.
All are stating between 500 and 1000 p/e cycles, even by Toshiba themselves ; not higher. Pretty sure endurance is higher now, at same shrink node, but, they are going onto 10 and 7nm (even though it's not real 7nm, but that's not the point).
TLC even with 3D-NAND gets less and less reliable every time they shrink. And that also why Intel is strugelling right now.

I'm quite afraid of an 256GB SSD with QLC...1TB you can see it coming, but 256GB or less ...

That's for general purpose processors, not NAND or even DRAM. Also as sirlesliechao said you're probably mixing endurance with reliability. We haven't had any endurance tests for QLC & endurance tests for certain regular MLC (non 3d NAND) drives proved that the TBW given by SSD makers are generally very conservative. Yields have nothing to do with either.

UpgrayeddIsn't SLC the most reliable? I thought the more levels meant higher capacity at a cost of reliability?

SLC is the most reliable.

The more layers are added, the more voltages have to be dealt with. This increases the variance/drift. It's why the Samsung 840, an early planar TLC drive, had to have its data regularly rewritten to the drive to prevent read speeds from falling.

In addition to the issue of how many layers there are (single = SLC, double = MLC, triple = TLC, quad = QLC) there are others:

1) process node size (larger is more reliable, e.g. 40nm vs. 15nm)

2) 3D vs. planar. 3D is more reliable. 3D fabrication is mainly what has made TLC so much more relevant in the market than it was initially.

3) various controller design factors, like DRAM-using vs. dramless (DRAM-using is better for reliability, except with Sandforce* which is a more complex case), error correction sophistication, and the presence of a capacitor to protect against power failure

4) how much NAND is set aside

5) the grade of the NAND

6) "SLC" caching. As far as I know, no drive has actual SLC NAND in its so-called SLC cache. Instead, a portion of MLC or TLC NAND is "functioning like SLC". This improves speed but I don't know how much it affects reliability. If it were true SLC then reliability would increase, certainly.

7) firmware/controller bugs

8) cooling (e.g. metal casing, thermal pads, heat sinks). This can affect performance consistency, which is a form of reliability (performance reliability).

*In terms of Sandforce, it was a dramless design that had improved theoretical reliability via on-the-fly data compression. This reduced the amount of data written quite a lot versus other controllers at the time; Sandforce had an excellent write amplication factor, with compressible data. Unfortunately, the design made TRIM mostly ineffective, causing the drives to slow, especially in steady state, and not recover. The design also wasn't very reliable (most severely with the Vertex 2 with 64-bit NAND in the 240 GB size) and was further worsened because Sandforce favored locking the drive to "prevent people from looking at their special firmware design". There were a lot of firmware updates for both the 2nd and 3rd generation controllers. It seems that the third generation eventually became pretty stable with 64-bit NAND but the performance, especially with a drive in need of TRIM recovery, wasn't competitive with later companies' designs. My speculation, as to the main reason why Sandforce is no longer a significant market player, is the poor compatibility with TRIM.

fynxer

YEAH.... THIS IS SO LIKE INTEL..... OH Fu**ing YEAH, Thats INTEL ALRIGHT!!!!

PS: C'mon intel your not exactly a forthcoming and honest company either>>>>