Optane / 3DXPoint - legacy and fate

[nageme]

Continuing a previous discussion about the discontinuation of Optane...

yet we're seeing new optane releases from intel (p5810x from 2023, for instance ...)

It's from a year before. Discussed/announced in early 2022, released late 2022.
Might have to do with Intel's enterprise-style longer-term obligations, or perhaps trying to use up all of its existing chip inventory.

They were gradually retiring also other Optane products, like DIMMs, which apparently might still ship until the end of 2025.

If you find any mention of what TI is doing or have done with the 3D XPoint fab in Lehi, the one they bought from Micron in 2021, do share.
A possible theory is that they're just using it as a general fab, that has nothing to do anymore with 3D XPoint. TI's announcements regarding the fab purchase didn't necessarily mention 3DXP.

 

[Toss]

I am happy that it's so good, and it will serve me 30-50 years easily,
I don't see any faster technology in the horizon.

M2 is laptop format which came to PC and crippling it with thermals.

 

[Selaya]

last information i have is that there was no wider mentioning of the lehi fab at all in press; basically every1 involved is doing their best in sweeping everything about it under a rug, which gives credence to the theory that TI is continuing the operation it as a 3dxpoint fab and selling its entire output back to intel

 

[LabRat 891]

Watching this thread; I Love the technology .
Recently had to move away from my P1600X 118GB Boot Drive. (AppData bloating.) 'Figured a 2TB 990 Pro wouldn't be much of a noticeable difference from the P1600X.
-I was wrong, and I have a 400GB P5801X+E1.S adapter(s) on their way.

I'd really like to know where 3DXpoint is going, as an IP.
Last I'd read, TI was interested in Optane for integration into neural networking hardware. -something about 3DXpoint having some inherit benefits for that kind of processing...

I am happy that it's so good, and it will serve me 30-50 years easily,

Sadly, there's other components on Optane NVME devices, much less-likely to last that long. The 'crosspoints' stored in the PCM though, will last eons.

I don't see any faster technology in the horizon.

There likely has been and is, right now.

Related "Ovonyx" memory has been in-use by BAE systems since before "Optane" hit the market. The issue, is that we (as 'consumers') will likely never see it.
"Optane" was a very special moment in the tech industry, and I don't think we'll see 'halo' or 'edge use' technologies touch the consumer market, ever again.

Ex: have any of you seen any Kioxia XL-Flash based drives for sale? Even used?

M.2 is laptop format which came to PC and crippling it with thermals.

Yes, it is. However, that is the compromise common to compact form factors (and mobile form factors).
-And, That is why M.3(short-lived), U.2/U.3, and E1.S/E1.L have become common for Enterprise SSDs.

 

[nageme]

Last I'd read, TI was interested in Optane for integration into neural networking hardware.

Where did you see that?
I got the sense that TI just bought a fab, not the technology or products that used to be produced there.

But maybe Micron is using some related low-level phase-change IP:

Micron has own take on novel selector-only memory

Micron Technology has developed its own version of a novel type of memory based on Ovonic Threshold Selector (OTS) devices.

www.eenewseurope.com

there was no wider mentioning of the lehi fab at all in press ... which gives credence to the theory that TI is continuing the operation it as a 3dxpoint fab and selling its entire output back to intel

If there's no indication it's producing any 3DXP, and there is some indication it's just using the fab as a fab, my guess would be that... it doesn't produce 3DXP, and it's using the fab as a general fab.

According to TI's page for the Lehi fab (previously known as LFAB, and now LFAB1) , it was purchased in 2021, and began production in 2022. So, a long downtime.

TI said this at the end of 2022:

Our company’s latest 300-millimeter wafer fab in Lehi, Utah, LFAB, has started production for analog and embedded products roughly one year after our company’s purchase of the facility.

LFAB has the capability to support 65-nanometer and 45-nanometer technologies with the ability to go beyond those nodes as required, and has optimal process technology to produce complex devices like embedded processing chips.

And in 2023, started to build LFAB2:

LFAB2, which will connect to the company's existing 300-mm wafer fab in Lehi. Once completed, TI's two Utah fabs will manufacture tens of millions of analog and embedded processing chips

Figured a 2TB 990 Pro wouldn't be much of a noticeable difference from the P1600X.
-I was wrong

What's the big difference in daily use?

 

[LabRat 891]

Where did you see that?

IIRC, It was in the forum discussions on the early drafts of the (now well-circulated) "There Is Nothing Wrong With Optane And 3D Xpoint" article. I believe, it was the author's 'informed conjecture', and that tidbit (amongst many) were omitted from the article, throughout its drafts.

I got the sense that TI just bought a fab, not the technology or products that the fab used to produce.

I got that feeling from what I'd read about the transfer, too.
BUT, there had been past 'chatter' about TI's (and Sun's, IBM's, etc.) interest in new, neurocomputing-focused memory technologies.

But maybe Micron is using some related low-level phase-change IP:

Micron has own take on novel selector-only memory

Micron Technology has developed its own version of a novel type of memory based on Ovonic Threshold Selector (OTS) devices.

www.eenewseurope.com

Micron Technology Inc. (Boise, Idaho) has developed its own version of a novel type of memory based on Ovonic Threshold Selector (OTS) devices.

Micron discussed the technology, which it calls Single-chalcogenide X-Point Memory (SXM), in an invited paper at the International Electron Devices Meeting held in December, 2023. The paper is entitled Status and Perspectives of Chalcogenide-based Cross-Point Memories.
The technology reported on is similar to developments at SK Hynix, dubbed selector-only memory (SOM), and at Samsung, called self-selecting memory (SSM), which were presented in the same session at IEDM (see Selector-only memory gains advocates, including SK Hynix).
The selector-only memory (SOM, SSM or SXM) is based on chalcogenide phase-change material in Ovonic Threshold Selector devices (OTS) that were used in 3D XPoint phase-change memories and that are used for various ReRAM devices.

New branch(s) on Stanford R. Ovshinsky's technolgic legacy?! Now, that's exciting!

What's the big difference in daily use?

Random Read/Write speeds, Access Times, Transaction Latency, etc. IoW: Just about every action a user might do in an OS that 'wakes up' or otherwise 'pages' C:, is much snappier.
To be totally fair... The difference may be unnoticeable to most normal users, otherwise used to 'high end PCs'. But, I think anyone would notice it, side-by-side. Most, would still say it's not worth the cost, regardless.

 

[Wirko]

I see everyone thinks of legal obstacles first but what about technological ones? 3D XPoint is decidedly 2D. Green and yellow blocks are/were built in two layers of storage cells, with no known plans of building out in the third dimension. There's a high probability than Micron and Intel never found a way to do that. Actually, many chip manufacturing technologies can't scale to 3D as of yet (dynamic RAM, static RAM, CMOS logic), and researchers only have vague ideas about how to do it.

Ex: have any of you seen any Kioxia XL-Flash based drives for sale? Even used?

Same question about Samsung 983 ZET SSD.

Also I just found a mention of SK Hynix in association with phase-change memory research in this article from 2022:

Intel has Optane chip hoard with no plans to develop tech – Blocks and Files

Intel has no 3D XPoint manufacturing capability, no plans to start XPoint manufacturing, and no commitment to develop XPoint media.

blocksandfiles.com

 

[nageme]

3D XPoint is decidedly 2D.

I'm not sure:

Intel 2nd Generation XPoint Memory | TechInsights

www.techinsights.com

But even if it were 2D, it was already useful in gen1. And it's not like capacities were too small to be useful.

 

[Psychoholic]

The only Optane drive i have had was an old 280GB Optane 900P Drive.
I bought it when they first came out and it came bundled with a star citizen key, lol
I remember it being snappy as hell for a boot drive.

 

[R-T-B]

last information i have is that there was no wider mentioning of the lehi fab at all in press; basically every1 involved is doing their best in sweeping everything about it under a rug, which gives credence to the theory that TI is continuing the operation it as a 3dxpoint fab and selling its entire output back to intel

My most recent drive has a manufacture date of December 2023 so I am inclined to agree.

3D XPoint is decidedly 2D.

It's decidedly not, as mentioned, but its very density challenged due to scaling costs atm anyways.

and there is some indication it's just using the fab

Is there? All I hear is silence since the purchase. I'd say we have no info either way, as well as some suspicious assembly dates.

Sun's

I assume you mean Oracle? Sun has been gone for around a decade now.

 

[LabRat 891]

I assume you mean Oracle? Sun has been gone for around a decade now.

I did. Bad memory

 

[nageme]

My most recent drive has a manufacture date of December 2023 so I am inclined to agree.

Let's say production stopped in 2021. They apparently plan to fulfill orders until the end of this year, but if we call it 2026, that's 5 years of existing inventory.
Inventory is probably raw chips, and the production date is likely the complete assembled product.

From a 2022 article:

Intel reportedly had several years of inventory on hand, perhaps helping explain the magnitude of the $559 million write-off.

Intel will also continue to sell its other existing Optane data center products and predicts that, based on its current forecasts, it has several years' worth of inventory.

I did. Bad memory

No ECC.

 

[sause]

Found out about these drives in late 2024. Fallen completely in love. Currenty have a P1600X 118G, P5800X 400G and a P5081X 800G (ES).
They get 700MB 4kq1t1 and 5.5us latency in linux using ext4/xfs which I still marvel over.

Everything just feels so much snappier using these drives.

 

[LabRat 891]

Found out about these drives in late 2024. Fallen completely in love. Currenty have a P1600X 118G, P5800X 400G and a P5081X 800G (ES).
They get 700MB 4kq1t1 and 5.5us latency in linux using ext4/xfs which I still marvel over.

Everything just feels so much snappier using these drives.

'Been in love w/ Optane since I 1st got my hands on a <$20 M10 16GB Gen3x2 drive. In any USB-NVME enclosure, it's the absolute fastest OS Install drive I've ever used. Live Drives, even over USB 2.0 *feel* noticeably better, too.
-So much so, I kinda-sorta accidentally manipulated the market on those M10 drives.
Over a few months time, I ended up buying a couple-few dozen @ <$4/ea. Now, it's hard to find them for under $10-20. Did you know you can run a board out of addressing, even these days?!
Ooops...

I *just* got a P5801X 400GB (ES too, I think). Absolutely astonishing how much faster *feeling* the P5801X is over the 990Pro
(heck, even the 118GB P1600X, prior; I *felt* the downgrade in booting off the 990Pro)

My housemate thinks I'm nuts for my obession w/ Optane (he's probably right) *but*,
after I gave him 2 of my P1600X 118GB drives (for boot+Beam.ng appdata), he can't deny the performance.
Just, that the cost (and adapter madness, in the case of U.2 and E1.S Optane) seems unreasonable. Which, is fair.

 

[Hardcore Games]

I used to use a 64GB Optane but my Dell laptop has 64GB RAM which is far faster.

 

[LabRat 891]

I used to use a 64GB Optane but my Dell laptop has 64GB RAM which is far faster.

Can't boot from a (software) RAMdrive, AFAIK.

I think...
the UEFI would have to have a module to support a RAMdrive
or
another bootloader/OS would have to set up the RAMdrive. (Which, would negate 1 plus in comparison with Optane-as-boot)

 

[Wirko]

But even if it were 2D, it was already useful in gen1. And it's not like capacities were too small to be useful.

It's just that the price per GB was, and remained, too high for nearly every application.
Lots of info in the TechInsights article anyway, thanks for the link (but it's not dated unfortunately, so it's unclear when the information was made public).

It's decidedly not, as mentioned, but its very density challenged due to scaling costs atm anyways.

All true. Yes I was aware there were 2 layers (didn't know about 4). Let's call it 2.2D... but no more than that. Scaling would be either too costly (many steps for each additional layer) or impossible using known technology. In contrast, NAND layers are formed about 100 at a time on a wafer, not one by one, becuase the storage cell has been made simple enough.

 

[sause]

'Been in love w/ Optane since I 1st got my hands on a <$20 M10 16GB Gen3x2 drive. In any USB-NVME enclosure, it's the absolute fastest OS Install drive I've ever used. Live Drives, even over USB 2.0 *feel* noticeably better, too.
-So much so, I kinda-sorta accidentally manipulated the market on those M10 drives.
Over a few months time, I ended up buying a couple-few dozen @ <$4/ea. Now, it's hard to find them for under $10-20. Did you know you can run a board out of addressing, even these days?!
Ooops...

I *just* got a P5801X 400GB (ES too, I think). Absolutely astonishing how much faster *feeling* the P5801X is over the 990Pro
(heck, even the 118GB P1600X, prior; I *felt* the downgrade in booting off the 990Pro)

My housemate thinks I'm nuts for my obession w/ Optane (he's probably right) *but*,
after I gave him 2 of my P1600X 118GB drives (for boot+Beam.ng appdata), he can't deny the performance.
Just, that the cost (and adapter madness, in the case of U.2 and E1.S Optane) seems unreasonable. Which, is fair.

That usb-nvme enclosure sounds like a great idea! Might have to use that for the P1600X.

Congrats on picking up the p5801x. They were being sold at really good prices recently compared to what they normally go for.
Adapters are still lacking for the e1s format. Can technically just throw a u.2 adapter on it and slot it into just about anything at that point.

I bet your housemate must be ecstatic after trying it out for the first time. That was very nice of you to let him experience what these drives are about.

Edit: Heres that linux result on the 400GB drive. Raising the clock speed of the processor still raises the 4k random performance. Blows my mind when I see it. Guess theres still more left on the table?

 

[Wirko]

'Been in love w/ Optane since I 1st got my hands on a <$20 M10 16GB Gen3x2 drive. In any USB-NVME enclosure, it's the absolute fastest OS Install drive I've ever used. Live Drives, even over USB 2.0 *feel* noticeably better, too.

What does CrystalDiskMark say about this combination? In every review of external SSDs I've seen so far, NAND flash-based of course, the USB connection almost brings the data transfer to a halt when testing random 4K Q1 reads (less so with writes). The USB latency must be the culprit, otherwise I don't know what could be. See this review at Anand's for an example, and external enclosures for M.2 SSDs give similar results. Now can Optane over USB be totally different?

 

[Assimilator]

What does CrystalDiskMark say about this combination? In every review of external SSDs I've seen so far, NAND flash-based of course, the USB connection almost brings the data transfer to a halt when testing random 4K Q1 reads (less so with writes). The USB latency must be the culprit, otherwise I don't know what could be. See this review at Anand's for an example, and external enclosures for M.2 SSDs give similar results. Now can Optane over USB be totally different?

I'm pretty sure this is placebo effect.

 

[R0H1T]

What legacy of a failed product?

Spoiler