Comments Locked

70 Comments

Back to Article

  • SarahKerrigan - Thursday, July 28, 2022 - link

    Cool technology that never found a niche.

    Pmem is coming, one way or another.
  • Kurosaki - Thursday, July 28, 2022 - link

    Yeah. Not surprised as it could not compete in prise, or storage volume. And being fast meant being proprietary in most cases.

    Stupid setup, doomed to fail.

    If they just shelled out some nice 2TB drives with insane speeds, filling up the pciex5 x4 bandwidth, for the price of a similar Samsung SSD, with no lock-ins to the Intel platform, those drives would have sold like I ecream in Sahara.

    Something went wrong, most probably production pace. To slow manufacturing always lead to higher consumer prices in the end. The factory has to pay of somehow. I guess. But how fast?
  • Eliadbu - Thursday, July 28, 2022 - link

    It is not about proprietary, it's about not being competitive enough with NAND. Price is a huge key factor in storage space and with 200+ layers TLC/QLC NAND coming the performance could not justify the price. I guess they simply could not forge a way for it to be competitive in volume and price. Without clear way of price improvement and losing money it was time for them to stop betting on that horse and focus on other fields (like logic chip manufacturing).
  • zanon - Friday, July 29, 2022 - link

    >"It is not about proprietary, it's about not being competitive enough with NAND"

    But that's directly related to it being proprietary. Can't get competitive enough on pricing without massive volume production and a diverse industry behind new tech for a product like memory. Intel keeps trying to replicate their effective-x86 monopoly (technically duopoly but has allowed Intel dominance for much of the period) but that was an unusual set of circumstances that not merely do they not have the capability for the rest of the industry is actively wary of. If Intel had cheaply and widely licensed Xpoint from the start, including getting massive players like Apple onboard (who are very willing to pay a premium for performance including for storage) we'd be looking at a very different situation.
  • ATC9001 - Thursday, July 28, 2022 - link

    Agree....I remember reading about this 3-4 years ago or so and being so excited! Unfortunately, they couldn't bring the cost down....imagine having that persistent memory speed along with high-performance DRAM!...I was hoping these would be a bridge between high end SSD's and pure RAM :(
  • JohnLee-SZ - Friday, July 29, 2022 - link

    Check out DapuStor Xlenstor, it can replace Optane P4800x/P5800X perfectly.
    https://www.storagereview.com/review/dapustor-x290...
  • ballsystemlord - Saturday, July 30, 2022 - link

    It's not available. I just searched the whole web. You can't buy it.
  • JohnLee-SZ - Tuesday, August 16, 2022 - link

    It's not available in channnel market, but you can reach out sales@dapustor.com if you need it.
  • dgingeri - Friday, July 29, 2022 - link

    It's not so much it couldn't find a niche. It's that Intel was asking way too high a price for it. It could have made a great replacement for NAND, and was actually cheaper to manufacture, but Intel wanted too high a premium.
  • Oxford Guy - Monday, August 1, 2022 - link

    It wouldn't have been 'niche' if it had been priced more affordably.
  • grant3 - Tuesday, August 2, 2022 - link

    Lower the price so they could suffer bigger losses per unit at higher volume? Not even Intel is that bad at business.
  • sallgeud - Tuesday, August 16, 2022 - link

    In our case it was insanely affordable for our purposes. Nothing comes close to the levels of performance we get out of our Optane systems
  • ballsystemlord - Saturday, July 30, 2022 - link

    I also will miss optane's depature.
  • name99 - Monday, August 1, 2022 - link

    Tech that was never *allowed to* find a niche...
    For day one, Intel saw Optane not as a generic technology to be sold to all comers, to find its roles. No, Optane was supposed to be a differentiator to make Intel chips (and not all Intel chips, but Xeons, and not all Xeons, only super-expensive Xeons) appear even cooler. And when it turned out that Intel's vision for the few ways super-expensive Xeons could be paired with Optane didn't pan out, there was no plan B, because no-one else had been invited to the Optane party.

    This is basically the same story as Decadent Phase Sony (from the 90s onward) - some occasional interesting new tech, but always structured so as to (supposedly) boost some part of the ecosystem, rather than as standalone tech that could be used by any consumer electronics or computer vendor with a good idea. And we all know how Decadent Phase Sony turned out.

    It's not IMPOSSIBLE to sell hardware that's only relevant to your ecosystem. Apple does it. IBM does it. But it requires the company to have been built on this mindset, to have a realistic worldview of how much better their proprietary hardware needs to be relative to the extra price.
    If the business model you're based on is selling to all comers (eg both Intel or Sony) you probably do not have the skills to make that pivot.
  • Threska - Thursday, July 28, 2022 - link

    Memristor would have been a better thing to vest in.
  • michael2k - Thursday, July 28, 2022 - link

    Is that sarcasm? Because for all intents and purposes, Optane was memristor. It stores data based on the bulk resistance of the storage material, which is fundamentally what a memristor does; send a current at one voltage to change the memristor's resistance, send a lower current at a lower voltage to read the value of the memristor. Where an SSD stores data using levesl of 'charge', Optane and memristors store data using levels of 'resistance'
  • Threska - Friday, July 29, 2022 - link

    Except someone from Intel disagrees with you.

    https://www.theregister.com/2015/07/28/intel_micro...

    Optane is PCM.

    https://pcper.com/2017/06/how-3d-xpoint-phase-chan...

    The idea it being a memristor is debatable, even though Leun Chua and Stan Williams feels so.

    https://en.wikipedia.org/wiki/Phase-change_memory. Notes #6 and #7.
  • michael2k - Saturday, July 30, 2022 - link

    So it seems a true memristor has access to multiple levels, ie you can set the resistance anywhere between 0 and 1 while Optane can only set its resistance to two values.
  • hubick - Thursday, July 28, 2022 - link

    This is a huge shame. Optane SSD's were the fastest drives on the market for ZFS cache/log, running your OS, etc. Also, if you value reliability, with everything being MLC/TLC/QLC these days, Optane beats the pants of all those for longevity. I have four P4801X 100GB M.2 drives I will treasure until they die.

    Intel, you had the best product on the market, and you idiots FUBAR'd it.
  • bambinone - Friday, July 29, 2022 - link

    I was hoping to grab a couple P1600X drives but they never really came available at retail, at least not that I could find.
  • sallgeud - Tuesday, August 16, 2022 - link

    Seconded
  • Drumsticks - Thursday, July 28, 2022 - link

    Sad to see. If only from the prospect of an entirely new storage architecture actually making it to market, I was hoping Optane would do well. I picked up one of the original 280GB 900p's as a boot drive back when they were running the Star Citizen promo with a free ship that sold for nearly as much as the drive cost, and I love it. It's been carried forward through several PC builds now, and I'm hoping to keep it going for the foreseeable future.

    Hopefully we'll see something cool come out of another company someday soon.
  • NonSequitor - Friday, July 29, 2022 - link

    Same here - riding my 280GB drive for as long as I can. I might have to pick up one of the M.2 units as well just in case it becomes tough to hook this drive to newer motherboards.
  • xol - Thursday, July 28, 2022 - link

    The unsold inventory says a lot.

    Intel offered this as a niche product from day one, even though they claimed to be serious about it. With hypothetical hindsight they should have gone all in to practically destroy the conventional flash business (as a simply better product) not supplement it.. But that is just internet commentator talk.

    The investment needed to scale things up massively only comes from mass market, and they decided to skip that for high margins selling to data center etc.

    This story feels similar to magneto optical discs from way back.
  • Zoolook - Saturday, July 30, 2022 - link

    They went all in on RDRAM, that didn't work out so well.
  • Oxford Guy - Monday, August 1, 2022 - link

    It wasn't better than the competition. Optane was better than the competition in some key areas.
  • Duwelon - Thursday, July 28, 2022 - link

    "Sapphire Rapids, in turn will be Intel’s first CPU to support CXL, and the overall technology has a much broader industry backing."

    It's doomed to fail. Why does it require CPU support, what can't they do with general x86 or it's ten trillion extensions? I'm sure there's a "good" reason but it's doomed to fail all the same, make it generally available or don't make it all Intel.
  • Threska - Friday, July 29, 2022 - link

    Feels like they haven't learned anything from Thunderbolt. USB4 just may eat their lunch.
  • back2future - Friday, July 29, 2022 - link

    me don't think it's the whole story, since why would Intel discuss company strategies with their (revenue related) customers?
  • ArcadeEngineer - Friday, July 29, 2022 - link

    It requires CPU support because that's where the memory controller is, it has nothing to do with the ISA. You could say the same thing about DDR5.
  • SarahKerrigan - Friday, July 29, 2022 - link

    I think you're confused as to what CXL is. Hint: It requires new controllers - and by the way, it's new in AMD Genoa too.
  • throw-awway - Thursday, July 28, 2022 - link

    I've been working on 4rd gen (cxl based)

    Ehh, :(
  • throw-awway - Thursday, July 28, 2022 - link

    4th*
  • Kamen Rider Blade - Thursday, July 28, 2022 - link

    NOOOOOOOOOOOO!!!!!!!!!!!!!!!!!!!!!!!!!
  • Silver5urfer - Thursday, July 28, 2022 - link

    Optane was king of NAND technology. Intel did the SSD NAND way before Samsung and others joined. The investor greed simply did not care on how the company has been faltering and rode the ship. Once the 10nm disaster started, the greedy WallSt left Intel to rot and used it to short the stock and moved all assets and money to Apple. Look at how cheaply they sold off 5G to Apple.

    Once the NAND market exploded. Intel and Micron's 3D XPoint didn't deem worthy to these investor pigs because ? Simply higher R&D and more money required for that niche market. And consumer class Optane was already super rare and barely used except for Caching and etc.

    The Optane tech in Enterprise however was top notch, P5800X is the undefeated and probably no one will ever catch it. The Endurance is insane and supreme performance with zero throttling or any other nonsense of speed drops like all PCIe NVMe SSDs was truly remarkable. Shame it was too expensive and virtually hard to run on traditional motherboards due to absence of U.2 connectors, we can get PCIe to U.2 anyways but it's a head ache, only EVGA offers U.2.

    Truly end of an era. This is what Investor greed does. NAND is very high profits and low R&D vs Optane. I wish we could procure a bunch of P5800Xs but man it's unobtanium level.

    Well now the only hope lies with Micron NAND. At-least in the current market their endurance is highest and performance as well. Samsung also dropped ball with dropping MLC and no replacement with their TLC.
  • ArcadeEngineer - Friday, July 29, 2022 - link

    Optane isn't NAND.
  • Silver5urfer - Friday, July 29, 2022 - link

    I meant memory, my bad.
  • JohnLee-SZ - Friday, July 29, 2022 - link

    Check out DapuStor Xlenstor, it can replace Optane P4800x/P5800X perfectly.
    https://www.storagereview.com/review/dapustor-x290...
  • ballsystemlord - Saturday, July 30, 2022 - link

    It's not available. I just searched the whole web. You can't buy it.
  • beginner99 - Friday, July 29, 2022 - link

    Very sad. Was very interesting tech. Hard to tell if it was actual production issues and cost or intel greed that destroyed the product or both. The cost was never justifiable to me over a normal NAND ssd. Maybe double the price could have been justifiable but not 5 or 10x times the price per GB.

    Since Micron also ditched it, I assume the production was a huge issue. But it's a chicken-egg problem. If you don't invest into R&D it won't get better.

    On top of this the probably even bigger issues is that software in general still hasn't really adjusted to the switch to ssds and must still work fine in hdds. All the speed and random access perfromacne don't help that much if the software intentionally is made for slow access speeds.
  • r3loaded - Friday, July 29, 2022 - link

    They never released a consumer follow-up to the 905P which was soon overtaken by TLC SSDs. Hard to buy in to the technology if it's not made available. If they'd come out with a new gen PCIe 4.0 SSD with random performance into the millions of IOPS it would have sold like hot cakes even at a premium price.
  • TrevorH - Friday, July 29, 2022 - link

    It's lucky that https://ark.intel.com/content/www/us/en/ark/produc... doesn't exist then
  • mode_13h - Friday, July 29, 2022 - link

    @r3loaded did say "a consumer follow-up". That P5800X is even more than "a premium price", which is probably more like the 2-3x markup that you see with top-end consumer M.2 drives.

    Don't get me wrong - I love the P5800X, but I would never spend that kind of money on one. I *would* have bought a PCIe 4.0 version of the 905P that used 2nd Gen Optane, if it were in the same price bracket as the 905P.
  • Tomatotech - Friday, July 29, 2022 - link

    Shame it never took off. So many baffling market decisions that it was like a man with both hands tied behind his back entering a boxing contest.

    IIRC as well as being high end, they also tried to sell it into the ultra-low-end, there were cheap laptops that advertised themselves as having something like 20gb of RAM - when it was actually 4gb ram + a 16gb stick of Optane. That caused bad feeling as the marketing was misleading - and bafflingly Intel never released these cheap sticks of Optane to the market even though millions of tech nerds like me were dying to get their hands on some Optane.

    On that point, given there’s apparently piles of unsold stock, anyone know if it will be sold off cheap? I still want some! :)
  • thestryker - Friday, July 29, 2022 - link

    It seems like the future of consumer storage has died with the Optane business. It certainly wasn't viable price wise on first gen, but it was close. NAND prices plummeted, Intel pivoted to memory, and killed off consumer Optane. Had they stuck with 3dxpoint for another generation they easily could have gotten prices close enough to NAND SSDs to carve a premium market. This really seems like a situation where investors don't like losing money, and Gelsinger isn't interested in being in the memory business.

    This is all a shame because the P5800X is the best persistent storage device on the market hands down, and NAND will never get close to it. I hope consumer level storage will get something better than NAND, but it seems extremely unlikely with the enterprise shift to CXL.
  • JohnLee-SZ - Friday, July 29, 2022 - link

    Xlenstor from DapuStor, same as P5800X.
    https://www.storagereview.com/review/dapustor-x290...
  • thestryker - Friday, July 29, 2022 - link

    AFAIK that's based on SLC NAND which can basically never get serious price reductions unless there's a huge breakthrough. It's the same issue Intel had with the first generation Optane, and while less so still there with the second generation. Another doubling of the layers would have allowed them to sell them more broadly for less, but I don't believe the same can be done with CL-Flash.
  • thestryker - Friday, July 29, 2022 - link

    *XL-Flash

    Sigh, tablet changed it and no edit buttons still.
  • JohnLee-SZ - Tuesday, August 16, 2022 - link

    3DXpoint is unique media and mask expenses of manufacturing rises a lot when scale up the layers, but XL-FLASH can leverage the NAND manufacturing process and more easier for scale up, then the cost will go down.
  • JohnLee-SZ - Friday, July 29, 2022 - link

    Optane/3DXpoint technology should be focus on enterprise / datacenter storage. But the new PCM takes too much money to expand for lower cost, with too many masks and tow low stack (2 & 4 stacks) .
    Check out DapuStor Xlenstor, it can replace Optane P4800x/P5800X perfectly.
    https://www.storagereview.com/review/dapustor-x290...
    Based on the optimized SLC (KIOXIA XL-FLASH), much easier for saling up and lowering the cost.
  • edzieba - Friday, July 29, 2022 - link

    A shame to see the Chalcogenide process die such an early death, after decades of Micron trying their hardest to prevent it even getting off the ground the first place (all the shenanigans around the Ovonics patents and avoiding paying royalties, and waiting until ECD could be killed off before unveiling 3D X-point).
  • zanon - Friday, July 29, 2022 - link

    That's a real shame. I wonder how much of it was due to typical reviews and advertising completely failing to highlight what a difference Optane was. The normal benchmarks tend to be high queue depth and large blocks, and no indication of 95/97/99/99.9 percentile latency spikes. I've tested supposedly top tier NVMe "very fast" drives and seen the performance just tank miserably in a variety of workloads which don't match very well. I don't think the price differential had to be as high as it was if it had scaled enough.

    I do think Intel/Micron mismanaged the tech though, not an uncommon pattern for Intel at all. They needed more partners and production, to get it to be something lots of places could produce and use standard and get at most a small royalty. Intel wanted their typical vertical lock-in which meant it'd only ever be ultra high end which meant lower volume and death spiral commenced. Stinks though, I'd always hoped one day that kind of consistency would be there for us all in volume.
  • mode_13h - Friday, July 29, 2022 - link

    Do we really think 3D XPoint simply lacks the potential ever to compete with NAND in bits/$? Or was Intel simply unwilling to invest the money to make it competitive, due to its poor market performance (which seems largely due to far fewer bits/$)?

    Also, I think the article could've focused more on the amount of software investment required (by Intel & others) to truly take advantage of Optane DIMMs.
  • thestryker - Friday, July 29, 2022 - link

    With CXL coming their hard shift to Optane DIMMs rather than expanding the storage side is baffling to me. Second generation 3D XPoint had twice as many layers as the first, and if they could have doubled it again the possibility of getting storage down in price to a reasonable premium over NAND was there.

    It really seems like a combination of shifting the market wrong and not wanting to spend the money to continue the project.
  • JayNor - Saturday, July 30, 2022 - link

    The Aurora project reportedly includes Optane. It isn't clear if it was waiting for gen3 Optane with a DDR5 controller, or if was to be integrated into some DAOS storage system with gen2.

    There were other testimonials by BSC, Meta, Alibaba that said it was the only viable solution for their HPC and storage needs. I believe for BSC it was the power vs DDR.

    The initial performance leaked results from the gen3/DDR5 indicated it wasn't ready. I wonder if now they'll remove Optane support from Emerald Rapids.

    There were also reports of new Optane packaging associated with the cxl memory pools that could have made an interesting product.

    You'd think with the wide ucie and Intel's advanced packaging plans, they could have found an co-packaged chiplet solution that improved the yields.
  • hanselltc - Monday, August 1, 2022 - link

    Optane as memory instead of storage always seemed a dumb idea to me. WTF is persistent memory good for when the systems it would've went into are made to have as little down time as possible? What good even is persistent memory, moving stuff from ssd into ram is not that slow, while optane compared to ram was always way slower. On storage it had a future, on memory it might have outside of DC, but intel went headfirst down the goldengate.
  • back2future - Monday, August 1, 2022 - link

    one financial reason, cost/GB (additional level 3-4) memory, maybe (previously) availability, compatibility with system changes, and some kind of 'out of system' backup with hot plugging
  • name99 - Monday, August 1, 2022 - link

    Your response shows why Optane was doomed from the start -- because people inside Intel, like you, can't see beyond the server.

    Imagine something like Optane in an Apple watch, serving as both DRAM and storage. Apple watch can live with the slower access because it has large caches, and it appreciates both the smaller package size (just one chip, not DRAM and flash) and lower energy/persistence compared to DRAM.
    But that would have required
    (a) Intel being willing to see this as not just as Intel-Xeon-Platinum exclusive AND
    (b) Intel being willing to make whatever changes Apple would require to make it practical (like maybe a smaller chip to match Apple watch's limited storage).

    Optane died because the people in charge of it were convinced the only use case that mattered was running SAP in-memory, and were too damn stupid/ignorant to understand that most computing has moved way beyond that sort of world!
  • Tomatotech - Tuesday, August 2, 2022 - link

    Interesting idea about Optane in an Apple Watch (or iPhone or other mobile device). I certainly never thought of that.

    However, Apple almost certainly has looked into it. They have trillions in cash and a history of scooping up any tech that can bring even a minor edge to their devices. The fact they haven’t put Optane into their devices suggests they found it brings little to no benefit. That could be in terms of power consumption vs speed, even on the high end, or whatever their criteria is.
  • name99 - Wednesday, August 3, 2022 - link

    Apple HAVE looked into it. There are a bunch of Apple patents about how to modify aspects of the SoC to work with a RAM that's slower than normal DRAM. (The patents don't say Optane but they are about a RAM that's slower than DRAM and persistent; my guess is that the work was generic to cover Optane and/or MRAM and even Nantero if that ever becomes real.)

    But Intel had their heart set on Optane being a "differentiator" that would force you to buy a Xeon Platinum because that was the *only* way you could get to use Optane DIMMs...

    It's certainly possible that Apple's conclusion after a few years of work (the patents span about three years) was that they could not get it to work, either technically or financially. But I suspect it's more likely that Intel was just not willing to play ball at first, and by the time they perhaps were willing, by crunch time, Apple was no longer interested. After all why should they be? Wait a few years and MRAM will fill the same niche (for Apple, anyway) without having to put up with Intel's prima donna attitude.

    Why do I blame Intel? Well look at the point I made – Optane is potentially viable in many UNUSUAL use cases. But Intel did nothing to even attempt to cultivate these. Compare with, eg, Toshiba's 1.8" drive. Toshiba had no idea what such a small and slow drive might be useful for, but they shopped it around and Apple bit, using it for the first iPod.
    https://whynow.co.uk/read/the-ipod-died-lets-bring...
  • deil - Monday, August 1, 2022 - link

    I still think they missed opportunity, like all storage would benefit from dedicated drive map on optane, both hdd's and ssd's would, 2/4/8GB of octane not as SSHD "cache" but dedicated high speed drive map, so hdd would save operations for file content's effectively tripling usual performance and ssd would get 1M IOPS from get go.

    Optane was at least useful tech, which had both promise and future, and it is dead only because of intel overconfidence....
  • six_tymes - Monday, August 1, 2022 - link

    I think they are making a mistake. they should have released second gen and see how the market reaction for another 2-3 years. YES I am aware they were loosing money, but the product is one of the best they have ever produced, and it needed more time in the market for uptake, espcially in server space.
  • escksu - Tuesday, August 2, 2022 - link

    ITs a good technology but ultimately failed.

    IF there is 1 single reason for its failure. Well, its simply not good enough. Remember the claims of it being 1000x faster than NAND? Well, we never did see that even till today. So, for its price and hype, its just didn't show that level of performance vs NAND.

    Of course other factors matters as well. But I consider this the biggest factor. If its even 10x faster (don't talk about 1000x), it could have been an incredible feat...
  • sallgeud - Tuesday, August 16, 2022 - link

    We've got a scenario internally where it's over 150x faster. That specific sceneario is enough for us to buy-up every single remaining one we can find. This stuff was like magic gold to us. It dominates in our specific use case. Might be why it died. Poeple like us who can make use of a ton of super-high performance disks < 1TB in size are rare.
  • six_tymes - Sunday, August 7, 2022 - link

    what a shame. too good for people to understand and embrace it.
    the lower quaility wins. sad.
  • Bruzzone - Monday, August 15, 2022 - link

    Phase change was doomed from the start. I was there with Jeff Handy from Objective Analysis attempting to use my Intel production economic data to determine when phase change would yield Intel a positive return on investment. My data was unnecessary. The answer from day one was never on continuing capacity for price performance increase of NAND. Phase change of essentially mylar to keep it simple I realize that's not exactly correct also has issues around heat subject assembly and in use that flips bits. There are superior solid state memory technologies looking to displace captive charge whether nonvolatile or volatile on the way to replacing DRAM. Solid state memory design producer's objective is to license their technologies to large scale manufacturers; resistive ram, tunnelling, spin, other ferro. These are more interesting solid state 'replacement' technologies then phase change, relied in industrial embedded applications now at Mb densities moving to GB is the question. mb
  • sallgeud - Tuesday, August 16, 2022 - link

    JFC NOOOO... We use them in our databases and they absolutely KILL IT. They dominate every other solution out there and make it possible for us to reboot the DBs in minutes instead of hours.
  • Maxxie - Tuesday, August 30, 2022 - link

    Optane as memory enhancement has been a losing proposition from day 1. It clouded the market for the vastly superior dedicated solutions. I use the 280p storage an i7-8700K desktop at 1% low power mode, and it is so fast that I don't have much if any slowdown. It also runs cooler despite the higher power the storage uses, because the rest of the system on low power saves even more. I wish Intel would just combine this technology with their newer netbook processors - it would make chromebooks obsolete.
  • fuzzynyanko - Thursday, September 29, 2022 - link

    Optane in a DIMM form made sense for a server motherboard, but not the regular user. It was cool technology, even though I could never find myself to purchase it.

    I hope it gets licensed out. There might be still good use for the tech
  • stux - Sunday, October 16, 2022 - link

    A real pity. AFAIK nothing comes close to Optane for ZFS SLOG purposes.

Log in

Don't have an account? Sign up now