Comments Locked

72 Comments

Back to Article

  • nandnandnand - Thursday, December 24, 2015 - link

    2018? Gee, take your time WD/Seagate/SDK.

    All the technology needed to create a 256 terabyte 2.5" SSD is close at hand (V-NAND with TSV stacked dies). 16 TB 2.5" SSDs are available. HDDs have lost the maximum capacity war. That leaves HDDs clinging to their affordability lead in $/GB. HDDs at about $0.0225/GB ($180 for 8 TB), SSDs around $0.23/GB ($110 for 480 GB). The 10x difference will shrink rapidly as long as HDD pricing remains stagnant as it has since the floods.

    http://arstechnica.com/gadgets/2015/08/samsung-unv...
    http://www.anandtech.com/show/9520/toshiba-brings-...
  • melgross - Thursday, December 24, 2015 - link

    Take your time? You understand that this is bleeding edge technology, don't you? You think perhaps that they are being lazy?
  • Murloc - Thursday, December 24, 2015 - link

    no, he's thinking that the progress isn't fast enough to save their market from being eroded further by SSDs.
  • Solandri - Monday, December 28, 2015 - link

    NAND prices leveled off in 2013. Its Dec 2015 price is actually slightly higher than Dec 2014 price.

    http://blogs-images.forbes.com/jimhandy/files/2014...
    http://en.chinaflashmarket.com/

    This isn't to say it won't start dropping again. The memory industry tends to be cyclical. But if you're extrapolating SSD prices from their rapid decline during 2010-2012 to conclude that HDDs are dead, you're going to be disappointed.
  • Solandri - Monday, December 28, 2015 - link

    I should add that NAND is probably a dead-end in terms of technology. The problem is it works by holding an electrostatic charge in a small cell. When you shrink the size of the cell, the ratio of surface area to stored charge increases, and the charge leaks out more quickly (as a percentage) over the same period of time. Samsung already ran into that problem with the EVO 840 (19nm planar), and reverted to larger cells (40nm stacked vertically) for the EVO 850. But bottom line is moving forward, NAND is not going to benefit from advances in DRAM technology (mostly shrinking).

    If some of the competing nonvolatile memory technologies pan out, then we may see rapid drops in SSD pricing again. But I wouldn't hold my breath.
  • OCedHrt - Tuesday, December 29, 2015 - link

    That's what VNAND is for. And HAMR just sounds like Sony MD discs at higher density and multiplatter.
  • Kutark - Tuesday, December 29, 2015 - link

    ^ This
  • extide - Thursday, December 24, 2015 - link

    As long as HD's keep increasing in size, they will continue to have a significant lower $/gb -- it will be a VERY long time before SSD's are comparable $/gb
  • imaheadcase - Thursday, December 24, 2015 - link

    Not only that, but SSD prices are tied in with ram prices so makes them more volatile price wise. Which goes along with limitations for SSD when you can't get the high cap memory.
  • meacupla - Friday, December 25, 2015 - link

    Why or how does RAM prices tie into SSD prices, when they aren't the same thing at all?
  • FunBunny2 - Friday, December 25, 2015 - link

    NAND and [D|S]RAM are made on the same machines. sort of.
  • CaedenV - Saturday, December 26, 2015 - link

    Saying that they are directly tied to RAM prices is not exactly right... but the technologies are both very similar from a manufacturing standpoint, and an ability to improve yields or lower cost in one will typically bleed over to the other.
    Compare that to HDD manufacturing which is essentially the art of making rust dust store information... improvements there are almost entirely inapplicable to any other technological field.
  • FunBunny2 - Saturday, December 26, 2015 - link

    -- the art of making rust dust store information

    while I, too, like to call HDD spinning rust, the fact is other metals have been the standard for more than a decade. here: http://www.pcguide.com/ref/hdd/op/media_Media.htm
  • close - Saturday, December 26, 2015 - link

    All the technology needed to create a 256TB SSD out there, just really far away on the same shelf as the one needed to build a 256TB HDD. Just as you wait for denser chips for SSDs you wait for denser platters for HDDs. Just as you add more chips and PCBs in the same SSD case you add more platters and heads in the same HDD case.
    And there's one commenter here who even "invented" the 5.25" HDD that's a gazillion times faster that an SSD while storing a lot more. That's because we're talking about the "technology" that's out there.

    Also, have you read anything about the 6TB SSD from Samsung lately? It's not really on the market and even the price is speculation at this point. Most likely it's 50 times more expensive per GB than a normal HDD. The tech may be somewhere on a shelf but the truth is even in 3 years prices per GB for SSDs will still be 6-8 times higher then for HDDs by every analysis, even ones discussed around here.
  • bebby - Monday, December 28, 2015 - link

    The question is not about the price gap but about market demand: Even if a 6TB HDD is still cheaper, the consumer demand for more than 1TB HDD/SSD is very small in a time where everything is in the cloud. Demand for large HDD remains to be in the server space which is much less profitable than consumer. There is good reason that Samsung sold their HDD biz, because they understand this fully. BTW, SSD and DRAM price cycles are not the same as the price cycle is more about supply discipline and level of competition. There are only very few real competitors in SSD whereas DRAM is fully commoditized. SSDs could be much cheaper if the suppliers would be less disciplined.
  • ddriver - Thursday, December 24, 2015 - link

    Making magnetic platters will always be cheaper than making flash memory chips. There is a whole universe of possibilities for improvement in HDDs, however the market is deliberately not taking that route, instead keep on milking the same old runt cow.
  • close - Saturday, December 26, 2015 - link

    Also, if they'd just use 4" wide magnetic tape we would have storage devices capable of holding more than 1000 times what a normal HDD can store today with at least 10 times the resiliency and speed. I have all the calculations done. But the bastards in the industry...

    Back to you ddriver.
  • stephenbrooks - Thursday, December 24, 2015 - link

    Yes, HDDs aren't exactly "clinging" to a price advantage when that price advantage is a factor of 10. The NAND makers would have to do something pretty drastic to close that gap, even by the 2020s.
  • Samus - Thursday, December 24, 2015 - link

    That's the problem. NAND technology has had substantially more breakthroughs in a shorter period of time that magnetic technologies. The irony is magnetic technology now needs to be augmented with optical technology to continue increasing aerial density.

    Long story short, as the OP said, by 2018, 4TB+ NAND drives will be mainstream and the HDD industry will still be using SMR as a crutch to increase storage capacity at the expense of reliability and performance. The only saving grace will be Hitachi's helium drives, which will likely reach 16TB by then.
  • Lonyo - Friday, December 25, 2015 - link

    16TB SSDs are not "available", they have been demonstrated, with an expected release date of next year and a price tag expected to be over $5k. That's 2016 and over 10x the price/GB of mechanical drives.
  • CaedenV - Saturday, December 26, 2015 - link

    Yep. For bulk archival storage HDDs have a future, but outside of that, who would use it? Most companies just need 2-5TB of storage, with a far greater need for IOPS than greater storage. When looked at a $/IOPS standpoint (not to mention power consumption, reliability, etc.) the SSDs are much cheaper than HDDs.
    Still lots of market for cheap bulk storage though, so the HDD market is not going to disappear any time soon.
    Personally, I still use plenty of HDDs in my home server, that that will not change... but it is essentially a crime for anyone to offer a HDD as a system drive in a computer any longer. Even crap eMMC storage offers better user experience than a HDD.
  • Oxford Guy - Tuesday, December 29, 2015 - link

    Why not stop doing the spinning platter thing and have static platters with imaging sensors? Instead of the platter spinning the sensors would turn on and off. This would get rid of mechanical wear and would improve random access.
  • Hephastion - Thursday, December 31, 2015 - link

    Note that TSV stacked die are expensive. You are no longer increasing the number of die per wafer with process shrinks, but rather just increasing the number of chips used in a given product, and these chips are not cheap. Don't expect stacked die to result in the same cost savings as process node shrinks.
  • SeleniumGlow - Thursday, December 24, 2015 - link

    Well, will these be as fast as flash/ssd storage in 2018?
    I would think that hard disk drives would be the "Tape drives" in future, if they can retain data that long with proper enclosing. That ways, this tech makes sense. But it's best to wait for the economic side of this to come out.
  • DanNeely - Thursday, December 24, 2015 - link

    HDDs will never catch up with SSD speeds. Seek times for moving drive heads mean there's no chance of random IO not continuing to suck in perpetuity; and with spindle speeds capped for noise/vibration/power reasons sequential IO can only scale with increases in platter density.
  • squngy - Thursday, December 24, 2015 - link

    Seek times are irrelevant when it comes to multi media.

    If you want to store terabytes of 5k video an HDD might still be the best choice.
  • meacupla - Friday, December 25, 2015 - link

    If you don't care about money, then SSD can and does offer more density for a given size.

    For example, how many 1TB mSATA/m.2 drives can you fit into the same space as an 8TB 3.5in HDD?
    Well, it's a lot more than 8TB, that's for sure.
  • Klimax - Friday, December 25, 2015 - link

    Overhead will be killing...
  • PrinceGaz - Friday, December 25, 2015 - link

    How many 1TB mSATA/m.2 drives can you fit into the same space as an 8TB 3.5in HDD without them overheating?
  • patrickjp93 - Saturday, December 26, 2015 - link

    Sequential can increase quite a lot. Did you know current drives don't stripe data across platters? You can literally increase sequential performance by a factor of 2*platter count simply by implementing a new controller.
  • FunBunny2 - Saturday, December 26, 2015 - link

    I can't find confirming cites, alas, but lower brain stem memory tells me that mainframe drives of the 33xx vintage had both multiple (radial) head sets and bit parallel recording across platters.
  • melgross - Thursday, December 24, 2015 - link

    No. The rotational limitations, and that of the head can only be advanced so much. This is why 10,000 and 15,000 drives have never made it to the mainstream.
  • patrickjp93 - Saturday, December 26, 2015 - link

    WD Velociraptor says hello.
  • Gigaplex - Saturday, December 26, 2015 - link

    They were a short lived niche, I wouldn't really call them mainstream.
  • FunBunny2 - Friday, December 25, 2015 - link

    -- I would think that hard disk drives would be the "Tape drives" in future

    considering that most coders still think of data as a sequentially ordered byte dump, faster and bigger HDD is just what they want
  • DanNeely - Thursday, December 24, 2015 - link

    I'm curious what exactly "Other CE" is in the market segment share graph. HDDs in DVR boxes or game consoles? While I could see the former fading away with continued growth of streaming services; them being about a 5th of the total market today surprises me. I'm also a bit surprised they expect personal storage to hold on so well; that volume has to be external HDDs but as flash drives get bigger I'm seeing a lot less of those being used over the last few years.

    I'm also wondering why the 2.5/3.5" platter curve has platter densities converging. Are they expecting HAMR to just not work well on bigger drives or is it just a priority shift? The only reason I can think of for deprioritizing 3.5" development is the larger enterprise drive market share. As SSDs continue pushing enterprise HDDs to nearline/archival storage wouldn't most of the arguments favoring 2.5" drives over 3.5" fade out.
  • extide - Thursday, December 24, 2015 - link

    Yeah, I thought that convergence was a bit weird too. But remember that was only from Showa Denko.
  • FunBunny2 - Saturday, December 26, 2015 - link

    there are a couple of factors which could drive convergence:
    1) the working area of the two may well be quite similar. the inner cylinder of the 2.5 may be nearer dead center than the 3.5, while the outer cylinder farther, leaving both with approx. the same total area.
    2) ZBR implementation between the two types may tilt toward 2.5
  • name99 - Thursday, December 24, 2015 - link

    I'd guess much of "other" is surveillance/security.
  • chazzzer - Monday, December 28, 2015 - link

    DVRs account for a lot of drives. Virtually every cable/satellite household has at least one DVR, many have two or more.
  • The Von Matrices - Thursday, December 24, 2015 - link

    Remember, these are densities, not platter capacities. I see no technical reason why the 2.5" market should have lower densities than the 3.5" market, so all that is happening is that they are merging. All this indicates is that the latest technologies is being simultaneously deployed in both drive sizes at once as opposed to having the 2.5" drives lag behind the curve. 3.5" drives will still have the highest capacities due to their platter size and number of platters.
  • DanNeely - Thursday, December 24, 2015 - link

    Read the caption again, it's per platter density not density/inch. If the latter is fixed between the two platter sizes, 3.5" drives will have platters that are consistently 2x as large because they have 2x the area.
  • 3ogdy - Thursday, December 24, 2015 - link

    Faster, bigger, cheaper and more reliable SSDs coming next year.
  • Lonyo - Friday, December 25, 2015 - link

    And the year after and the year after and the year after...
  • Hul8 - Saturday, December 26, 2015 - link

    The developments in SSDs I've seen in the last 6 months have been at the low end, though. RAMless TLC SSDs that are slower and cheaper (just) than previous generation.
  • Demiurge - Saturday, December 26, 2015 - link

    SSD performance can be misleading in certain high-end cases... regardless of how faster, bigger, and cheaper SSD's get, they will never approach the reliability until a different physical medium other than NAND is cheap enough to replace NAND -- the error correction has always been a limitation and has repeatedly required increasing the bits dedicated to ECC. I thought FRAM and a few other promising theoretical technologies back in the 2007-2008 years, but it really hasn't gone anywhere for a variety of reasons. HDD's are still the mainstay because they are more reliable, cheaper, and can be faster than SSD's in high write environment (when used in large RAID arrays). Even perhaps more chilling is that RAM is cheap enough that caching is still a viable solution. There really is nothing compelling the high-end markets to switch other than a misleading performance specification that some of the big companies have already discovered problems using the the new Cloud-based market.

    Reference: http://www.zdnet.com/article/what-we-learned-about...
  • Michael Bay - Sunday, December 27, 2015 - link

    As if higher platter densities won`t require complex ECC algorithms.
    And in consumer sector, SSDs are already at lease as reliable as HDDs are.
  • DanNeely - Sunday, December 27, 2015 - link

    Yup. One of the big reasons why HDDs when from 512byte sectors to 4k sectors a few years ago was to keep the size of ECC codes under control. Without the larger sector sizes, ECC codes would've ended up eating most of the gains from the last few generations of platter density.
  • Arnulf - Friday, December 25, 2015 - link

    How is
  • Arnulf - Friday, December 25, 2015 - link

    How does this HAMR technology differ from magneto-optical drives of 80s and 90s? Is it operating on the same technical principle?
  • Lord of the Bored - Saturday, December 26, 2015 - link

    The same basic idea of using a laser to heat the media so it's less resistant to magnetic change is used, but the actual implementation is very different.
    MO disks were read with lasers, like a CD/DVD/BRD. These are still read as magnetic disks.

    Tangentally, I still wish MO disks had replaced floppies instead of CD-RW. I remember when the first MiniDisks hit in the early 90s how great a floppy they would make(and then Sony ignored the possibilities as their music division shot themselves in the foot repeatedly).
  • Michael Bay - Sunday, December 27, 2015 - link

    Between MiniDisc and those zipdrives, absolutely.
  • HerveS94 - Friday, December 25, 2015 - link

    Interesting to ee this new technology about HDDs, so i am looking forward to seeing new storage tdch being implemented into this, it also looks very promising.
  • SydneyBlue120d - Friday, December 25, 2015 - link

    Do You think it will be possibile to sse SSHD using HAMR + Intel Optane ?
  • Scoobmx - Friday, December 25, 2015 - link

    I believe it's superparamagnetic limit, not supermagnetic limit, but maybe you are referring to something unrelated.
  • DesktopMan - Saturday, December 26, 2015 - link

    Unless HDD speeds catch up to this storage increase they will still be left behind. It already takes way too long to rebuild RAID arrays based on 2TB or larger disks.
  • The12pAc - Sunday, December 27, 2015 - link

    I think this a non-issue anymore since storage controller (at least most enterprise-class) are controlling the placement of bit for redundancy instead of the traditional RAID controller. My Compellent SAN will allow for all versions of RAID depending on how much redundancy/speed the customer wants. This illiminates the rebuild factor.
  • Amazing2u - Saturday, December 26, 2015 - link

    2 things I would like to see happen in this decade:
    - Smart phones with a high performance CPU/GPU that will last more than a day on a charge
    - a hard disk drives with a high storage capacity that is RELIABLE!!!
  • stang725 - Sunday, December 27, 2015 - link

    If and when 450mm memory FABs come online, you might eventually see the bi-modal price per GB merge, but not without 450mm. Platter-based storage and NAND are too far apart in terms of their manufacturing inputs for them to ever be at the same price, outside of putting more chips on a wafer and maintaining the yields on a larger wafer. Semicon is already close to the bottom-end of how small you can go in terms of the patterning (light-source limitations) to hope for any huge gains in chip density anytime soon, at least not without a huge jump in FAB equipment costs. Also, a 450mm FAB would not be cheap and would have to built from the ground up (talking billions of USD). I would be happier with just faster NAND/mem controllers/interfaces... the need for the AVERAGE 2015 PC user for more than 4TB of storage is not really there, especially with cloud storage.
  • ppi - Monday, December 28, 2015 - link

    Cloud storage is not an answer

    1) Cloud storage providers need to store their data somehow as well
    2) Did you compare cloud storage prices vs. cost of NAS? Not pretty for longer than 1 year.
    3) Local storage will always be faster than any cloud solution. I would admit at certain point it can reach "fast enough" speeds.
    4) What "average" user needs quite develops over time. There was time I thought my 120 MB HDD is all I will ever need. Did you factor in 4K videos, 100 GB games, etc. in that figure?
  • bebby - Monday, January 4, 2016 - link

    1) this is true, but they do it more efficiently, using less drives and each of us (higher utilization)
    2) I did and to be honest for a small user like me, the cost of NAS is higher (not counting the trouble in maintaining it), 50 USD/y for unlimited data vs. 500-600 USD for the NAS without the energy needed. And then I also have to backup the NAS.
    3) This depends very much on the cloud provider, I have 1GBit up/down fiber speed, so that I end up with up to 50 MB/s over the internet currently.
    4) The great thing about cloud solutions like steam are that you can always delete and easily reinstall games once needed. No need to backup. And with Netflix my usage of my home library decreased a lot, I could even do without it now.
    I currently use both NAS (6TB as a raid 1) and cloud but probably will not buy another NAS any more. And my 2 1TB ssds have been the best investments ever.
  • jann5s - Monday, December 28, 2015 - link

    the 450mm wafer stepper line is completely frozen at ASML, there will not be any 450mm fab anywhere soon.
  • FunBunny2 - Monday, December 28, 2015 - link

    "If and when 450mm memory FABs come online, you might eventually see the bi-modal price per GB merge, but not without 450mm. "

    the 450mm wafer has been "real soon now" for years and years. without an output sink, fabs will continue to tweak 300mm. kind of like the end of the WinTel monopoly; without an ever more bloated Windows, Intel couldn't shift ever more advanced cpus. and vice versa.
  • Rocket321 - Tuesday, December 29, 2015 - link

    What good is cloud storage when all ISP's are moving to a capped data / overage fee model?
  • dragosmp - Monday, December 28, 2015 - link

    @Anton - nice to see you at anandtech! I really liked your work on xbitlabs until not too long ago
  • Glock24 - Monday, December 28, 2015 - link

    Bigger capacity is good and all, but what about reliability? Webber sube HDDs got over 1TB reliability has gone down the drain. Yes, there are some reliable HDDs, but overall they are worse than before.
  • Alexvrb - Tuesday, December 29, 2015 - link

    I for one am shocked that there's not a single MC Hammer gag posted.
  • wolfemane - Saturday, January 2, 2016 - link

    Can't touch this!
  • izmanq - Wednesday, December 30, 2015 - link

    there's enough problem due to moving parts, magnetic interference, now they want to add heat into the factor, they should investing and research more to SSD instead
  • andrewaggb - Wednesday, December 30, 2015 - link

    Power consumption? I have to think heating things up with lasers is going to use more power than a traditional hard disk. I suppose laptops don't really use hard disks anymore but I'm curious if these drives will draw considerably more power or not.
  • Mj@uk - Wednesday, December 30, 2015 - link

    your making me feel old......
    the first hdd i bought cost £1000000/TB
    the last one cost £28/TB
  • SanX - Sunday, January 3, 2016 - link

    Sounds like one more dead tech is emerging. How they plan to focus into 300 Angstrem diameter focal spot? We can barely focus into 500 A even with our X-ray lasers operating on 469A wavelength

Log in

Don't have an account? Sign up now