STORAGE

  • 06/19/2015
    12:30 PM
  • Rating: 
    0 votes
    +
    Vote up!
    -
    Vote down!

SSD Prices In A Free Fall

With the prices of solid-state drives expected to reach parity with hard-disk drives next year, are HDDs doomed?

Hard-disk drive vendors point to the higher price of solid-state drives as a reason to keep on buying hard drives, but as Bob Dylan sang, "The Times They Are a -Changin'." The advent of 3D NAND has become a game-changer for the storage industry by increasing SSD capacity and dropping SSD prices.

By packing 32 or 64 times the capacity per die, 3D NAND will allow SSDs to increase capacity well beyond hard drive sizes. SanDisk, for example, plans 8 TB drives this year, and 16 TB drives in 2016. At the same time, vendors across the flash industry are able to back off two process node levels and  obtain excellent die yields.

The result of the density increase is clear: This year, SSDs will nearly catch up to HDD in capacity. Meanwhile, hard drives appear to be stuck at 10 TB capacity, and the technology to move beyond that size is going to be expensive once it's perfected. HDD capacity curves already were flattening, and the next steps are likely to take some time.

This all means that SSDs will surpass HDDs in capacity in 2016. There’s even serious talk of 30 TB solid-state drives in 2018.

So what about SSD price points? In 2014, prices for high-end consumer SSDs dropped below enterprise-class HDD, and continued to drop in 2015. A terabyte SSD can be had for around $300. Moreover, this is before 3D NAND begins to further cut prices. By the end of 2016, it’s a safe bet that price parity will be close, if not already achieved, between consumer SSDs and the bulk SATA drives.

Learn more about the changing storage landscape in the Storage Track at Interop Las Vegas this spring. Don't miss out! Register now for Interop, May 2-6, and receive $200 off.


This will put pressure on hard-disk drive  makers to lower prices, but, frankly, they’ve used up most of the tricks to reduce cost and are already at single-digit margins for bulk SATA drives, so they don’t have much wriggle room.

With parity achieved in capacity and price, one has to ask whether HDDs will still be needed.  SSDs are blindingly fast in comparison. Typically, large consumer SSDs are 5x the streaming performance and 5000x the random read/write rate. With low operating power and very low standby power, SSDs are ideal for large archives, too.

Additionally, wear-out isn’t an issue with SSDs. Those two node uplifts in the manufacturing process add literally years to the device life, and the economics of 3D NAND allow for extra over-provisioning, making the write life of the drive well beyond its time in a data center. This is especially true for archived storage, where writing is at a much lower rate.

There clearly is an inflexion point in the use of hard drives coming. Once parity is achieved, the transition to SSDs will become a tsunami. This transition is already well along for so-called enterprise drives. With price and capacity already matched or exceeded by SSDs, 7200 RPM and faster HDDs will quickly fall out of favor.

As in any transition, there will be points of resistance. After-market HDD spares will continue to be sold, though upgrades and replacements will increasingly use SSDs, especially in servers. The volume reductions in HDDs will probably lead to some major fire sales, though. These will all delay the day the last HDD ships, but do not expect a tape-like extended demise, with 30 years of predictions of the end of tape countered by ongoing increases in tape capacity. SSDs and HDDs basically do the same thing and there’s no reason to have both.

Speaking of tape, the SSD archive appliance likely will cause the demise of that hallowed medium. Today’s interest is more for rapid access to data, as demonstrated by Google’s Nearline cloud storage and Amazon Glacier. An SSD basis provides the desired low power with instant-on performance. Tape-based Glacier takes two hours to recover the first blocks of data.

Likewise, DVD-type archive storage will need a magic trick or two to remain in the race. A terabyte of DVDs will cost more than a terabyte SSD and that isn’t including the DVD library unit.

If you were told that a BMW and a golf-cart were the same price, which would you buy? That’s going to be the dilemma facing buyers sometime in 2016, with SSD and HDD. I think I know your answer!

PS: We may have a similar discussion in 2021, when ReRAM, PCM or some other solid-state solution, mounts a challenge to flash memory. 


Comments

Memory Evolution

HDD's had a great run these last few decades, but they now look to be going the way of the vacuum tube. But, the reign of the SSD may not last as long. There is a device on the horizon called the memristor - a resistor that can change value and retain that value even when the power is gone. And, it's not just an item of academic interest - Hewlett Packard is spending a huge amount on the development of this new passive component.

Re: Memory Evolution

I think the erosion of HDD sales will take place quite rapidly. SSd are plug 'n play replacements but much faster, and as prices collapse, no-one will opt for an HDD when an SSD equivalent is available....that's over 95 percent odf the market.

The installed base may find their vendor lock-in to be a problem for a couple of years, as arrays built for HDD can only use approved HDD as upgrades. The solution to that is Software-Defined Storage and much more open platforms.

As for memristors, maybe they'll arrive soon..They've been "next year" for several years now...but thy'll have to overcome a very entrenched base of flash products. Moreover, using memristors likely means big OS and app changes, since the preferred format will be NVDIMM to take advanatage of the very high speed. The changes in software to make this happen are wide-reaching, from compilers and link editors to stacks to soft power fail recovery to apps themselves, with the app changes being major.

Re: Memory Evolution

It might, eventually mean even more radical change. A "flip-flop" is either on or off. A memristor might have 5, 10, or even more values. It could mean the beginning of the beginning of the end of the whole digital paradigm. It might also mean the eventual end of centralized storage for PC's and portables. However, the machine that HP is working on will, supposedly, only using it as a "1 or 0" device.

Re: Memory Evolution

R&D into SSD storage is responsible for the current state of the market in which SSD has become a serious contender with no other contenders in sight. However, I feel that the mobile device market might also get some credit as it created a market and production of SSDs on scale.

Re: Memory Evolution

Good point, Brian. The markets today are intertwined.

Not Enough Production Capacity..perhaps ever.

The problem is capacity. Even with 3D NAND we will not be close to replacing the sheer weight of enterprise HDD capacity. There are only 9 fabs currently producing NAND, and it would require another 20 fabs, at a cost of $6-$8 billion each, for a total investment in the area of $120-$160 billion, to replace the 250 exabytes of HDD storage produced per year. Most fabs are not of the greenfield variety, the huge upfront expnse is defrayed by simply converting to smaller nodes. This does boost the storage capacity per fab, but doesnt actually add in another fab. if the fab guys start building fabs left and right, which they will not, it will destablize the NAND market. Then it will turn into the same situation as the zero-margin DRAM game, and I think they all learned their lesson with that. My prediction is that the fab output growth will remain slow and steady so the major players will retain their margins. Is there demand? oh yes. Is there a NAND shortage looming? I certainly think so.

 The performance HDD segment is definitely in trouble however, but i think the large capacity HDD will live on for quite some time for cold storage. 

Re: Not Enough Production Capacity..perhaps ever.

Paul,

I don't buy into the lack of capacity idea anymore. The advent of 3D NAND has allowed older nodes to be reused, saving those billions in capital, while cranking up the capacity per chip by a factor of 32 or 64x. Even though processing a 3D wafer takes longer, the overall process is much more efficient using a metric of petabytes per hour.

This will close the gap on HDD significantly, whcih is why the predictions of price parity in 2017 are coming from all the flash vendors. From an HDD point of view, the only solution is larger capacity, but we've hit the wall on this for at least a while.

Meanwhile, the storage systems types are crunching the demand issue from a different direction. We will see a massive improvement in compression technology in the next 2 years. With compression ratios in the 6:1  to 10:1 range for general data, and even 30 percent file size improvement on JPEG, the demand side is not going to grow as fast in RAW terabytes, and by a major factor.

Deduplication will also impact capacity demand. Estimates vary widely depending on data types, but a further 3:1 saving is possible and in some cases 100:1 (virtual desktops) is achievable.

Any way you look at this, the supply gap isn't anywhere as great as you think!

Re: Not Enough Production Capacity..perhaps ever.

Jim,

You make good points, but it is also important to remember that deduplication and compression also apply to HDD storage as well. With the compression and deduplication ratios that you noted the price-per-GB of HDDs has the opportunity to be insanely low. Yes, SSDs can handle the compression and deduplication workloads with ease, whereas HDDs take a bit longer. For cold storage, these techniques and performance penalties are more than acceptable. In any case, with tiering the compressed data will simply be passed to the slower tiers.

Many of the projections that I have seen for price parity rely upon the current price of today's HDDs remaining stagnant, and not the price they will be in the future. HAMR will be costly for the manufacturers initially, but when they are producing 20TB HDDs the price will still be lower than current prices on a dollar-per-GB basis.

Nearly every source indicates that we are simply not producing enough capacity to store the worlds data over the next decade, even with both sources of storage combined. There is a looming capacity gap, and I personally do not feel that the million+ HDDs produced per day (by one vendor alone) are easily replaced. For the performance segment, however, there is no doubt that is inevitably going to be all-flash.

I think it will fragment into SSDs for performance workloads, and massive HDD repositories for cold data using object storage architectures. Of course, everyone has their opinion, and some are more bullish on flash than others are.

Only time will tell :)

Re: Not Enough Production Capacity..perhaps ever.

@Paul, you made an interesting point. At the one hand, it could turn out that as soon as price parity is achieved, HHD's production will exponentially decrease and SSD's production will exponentially increase. On the other hand, as you mentioned the growing demand of data storage in the world due to mobile devices, IoT devices, video storage/streaming and a greater number of individuals connecting to the internet, both HHD and SSD might need to exponentially grow to meet world demand.

Solid state changes the state of database applications

Solid state changes more than the balance of power between hard drive and Flash. It has the power to change the nature of database applications and big data systems. When all the data can be placed in the equivalent of random access memory, a different scale and type of application can be employed. We are on the cusp of major breakthroughs here.

Re: Solid state changes the state of database applications

Charles, you absolutely right1

Computer hardware generally is seeing the real benefits of COTS now....fast agile development of new ideas. Not only is SSD superfast, networks are going to 25 then 50GbE in just a couple of years, with 4x versions giving 100 and 200GbE trunk links, while DRAM is heading for terabyte/sec transfer rates.

Re: SSD Prices In A Free Fall

It's nice to hear you come down decisively, Jim - it would have been easy to write an article that said 'is SSD the death of HDD? Well, you know... maybe....'. I'm certainly not in disagreement that it's true, but I was a little surprised to hear that 3D Nand was coming so soon. Is that the case even for consumer drives, or are they a little farther away? You make some good points, Paul; the real tsunami Jim describes can only happen if the demand is there that very day that consumers/enterprises want to make the switch. If not, it could delay adoption a few years until someone's next refresh cycle... which could put a slight drain on the whole ecosystem Charlie is talking about. That's still just talking about delaying the inevitable, though.

I'm inclined to agree that price is king when the end result is the same (or, in this  case, better). The only snag I can see in the consumer market is a couple of not-so-savvy uncles calling their nephews and asking 'can I trust this newfangled Solid State Thing (SST)?' It's not a case where hobbyists will say 'nay' out of sticking to the tried-and-true - the hobbyists will be the first ones to switch! - and it's not a case like tape where there's some special enterprise security/compliance/etc. need to stick with HDD. Lasty, I can see why Amazon would want to name their cold cloud storage offering 'Glacier' (it's clever)... but shouldn't they have went with something a little sexier? Do people really want to buy Glaciers?

Re: SSD Prices In A Free Fall

zerox, I saw a 3D NAND 250GB consumer drive on the web this morning for $99.99 priced down $50 from the start of the year. As Holmes used to say, "The game is afoot!"

I also saw an article where NetApp said they had NEVER seen a drive reach wear-out in the field. That's a mainstream-class endorsement!

NO Price parity before 2020

Jim,

While it's possible that for DESKTOP applications SSDs will reach price parity with HDDs in 18 months as you predict that will not be true for data center applications for several years.  While I think that many users, including corporate users and the cloud savvy, would be better served by using smaller SSDs and external storage than large SSDs today that's not primarily a cost based recomendation.

You seemed very impressed that a 1TB SSD was available for just $350. On newegg today that would be a Sandisk X300 client device and competitive 1TB HDD from WD or Seagate would cost about $50.  While the SSD could be about half what an equivilent SSD would have cost last year the HDD is 1/7th the cost of the SSD. 

If we look at data center use cases the price differential is even greater. An Intel DC S3500 1.2TB SSD costs $1110 (I'm willing to call it $1/GB) where a 6TB Seagate Archive  drive is $240 or $0.04/GB a differnece of 25 times. Even if I use the Sandisk X300 as my cost model the price difference is almost 10:1.

Buy the way those SSDs that Netapp said they never saw fail they weren't Sandisk X300s or even Intel DC S3500 (the low end of Intel's enterprise line) but more like the Intel DC S3700 with better flash and more overprovisioning. The DC S3700 is $1200 for 800GB or 1.5 times more expensive than the DC S3500 on a per GB basis.

In the desktop market we'll reach parity because desktop users need only say $500GB and the cost of small disk drives stops falling once the desired capacity can be met with one platter. A 500GB disk costs $40 vs the $50 of a 1TB disk for just this reason. The minimum price of a disk has to cover a platter, head and spindle motor so small disks are more expensive per GB than large.  When the 512GB SSD hits say $100 intelegent users will abandon the HDD.

However in the data center we buy both flash for performance and spinning disks for capacity therefore we buy big disks that cost a lot less per GB especisally when you fingure in the cost of an array or server drive slot.

For the cost of SSD and HDD to cross at say 4TB as you predict SSD costs would have to come down 80%, a factor of 5, in 18 months AND hard drive technologies like bit patterned media and HAMR would have to fail so that hard disk prices didn't decline.  

If we assume that SSD prices fall at 40%/yr (a bit faster than historical rates) and HDDs decrease at 15% (A bit slower than historical rates) it's still 5 years before parity

It ain't gonna happen Jim

 

Re: NO Price parity before 2020

It's fun arguing this with the HDD side of the house! I've just seen an ad for a $250 GB SSD at $99.99! We are nearly at that magic $50GB for $200!

If we compare "slow "enterprise" HDD with blindingly fast NVMe-PCIe SSD, there's a price difference strongly in favor of the HDD, but look at the performance....2,000,000 IOPS versus 300. Give me a break, Howard. That's comparing a push cart with a Bugatti!

If we come down the tree a little, comparing those same "enterprise" HDD with quality consumer class SSD, the story changes. A $600 1TB HDD compares with a $330 SSD; and again, the SSD has 90,000 IOPS versus the 300 of the HDD. SSD win on both price and performance.

The argument that the consumer SSD isn't robust enough for a server use is baloney. See NetApp's belief they've never seen a wearout problem in the field, for instance.

The only place SSD misses out is in the pure price per terabyte play when a cheap consumer hard drive is measured against the equivalent SSD. At $330 that 1TB SSD is expensive compared with a $39 hard drive.

The problem hard drives face is that we are sabout to have a 64x or more increase in capacity per SSD die, without any large increase in cost or processing time. That translates directrly into cost of product, and as 3D NAND hits its stride in 2015, we'll see 1TB SSD hit much lower price points,

Frankly, SSD doesn't even need to achieve parity to win. A 90K IOPS, 3Watt 1TB SSD at $150 will be very attractive compared with slow-as-molasses $39 1TB HDD in the server market, from archiving use cases all the way thorugh to tiered secondary storage to all-flash arrays. There won't even be a discussion in the virtualized server local instance storage case!

Even the capacity issue is a loser for hard drives. SanDisk has 16TB units planned for next year...that will be a strong competitor for that 6TB Seagate unit.

Re: NO Price parity before 2020

Re: NO Price parity before 2020

Jim,

To address you point by point:

You made a specific prediction that SSDs would reach price parity with HDDs by the end of 2016. Since SSDs are already cheaper on a $/IOP basis your prediction must be about $/GB.

You then argued:

>>If we compare "slow "enterprise" HDD with blindingly fast NVMe-PCIe SSD, there's a price difference >>strongly in favor of the HDD, but look at the performance....2,000,000 IOPS versus 300. Give me a >>break, Howard. That's comparing a push cart with a Bugatti!

I agree we're not talking about speed I don't know why you're bringing up PCIe, I didn't, all 3 SSDs I mentioned in my comment were SATA. $/IOPS is irrelevant to your prediction because SSDs are already cheaper per IOP.

>>If we come down the tree a little, comparing those same "enterprise" HDD with quality consumer class SSD, >>the story changes. A $600 1TB HDD compares with a $330 SSD; and again, the SSD has 90,000 IOPS >>versus the 300 of the HDD. SSD win on both price and performance.

I don't know where you get $600 for a 1TB hard drive. NewEgg is selling the Seagate Constellation ES 1TB, Seagate's basic 7200RPM enterprise drive for $85 and the Western Digital RAID Edition for $99. So even at 1TB the HDD has a 3:1 cost advantage.

>>The argument that the consumer SSD isn't robust enough for a server use is baloney. See NetApp's belief >>they've never seen a wearout problem in the field, for instance.

You can't take a random internet article about NetApp having never seen an SSD wearout in the field and extrapolate that to "consumer SSDs are robust enough for the data center". That would only be a valid conclusion if NetApp used consumer SSDs but they don't. If NetApp has never seen an SSD wear out, a likely truth, it's because they've over-engineered the SSDs they do use to have more endurance than they expect their users to submit the drives to.

To use your car analogy that's like taking a data point that 99% of 20 year old Mercedes-Benz cars are still on the road and using that as evidence that Yugos or Trebants will last 20 years.  M-B builds better cars, therefore they last longer, seems a better conclusion to draw than all cars last 20 years.

Enterprise SSDs not only have higher endurance but also power failure protection. All SSDs store logical to physical location maps in RAM in the SSD controller. When power fails on a consumer SSD some of that data is lost and some data can be corrupted. An enterprise SSD has a power fail detection circuit and capacitors to power the dumping of this volatile map. This ensures data integrity and increases the cost. 

While I'm in agreement that people are more worried about wearout than they should be but I've most certianly seen consumer SSDs wear out in data centers.

>>The only place SSD misses out is in the pure price per terabyte play when a cheap consumer hard drive is >>measured against the equivalent SSD. At $330 that 1TB SSD is expensive compared with a $39 hard drive.

I compared the consumer SSD (the Sandisk at $300) to the enterprise HDD (a 6TB Seagate enterprise archive drive), that's worst case not the best case you mention. That's where I got the 10:1 current price delta. That's best case for your side. If we use a data center SSD, even a low end one like the DC S3500, the gap grows by a factor of 2 or more.

>>The problem hard drives face is that we are about to have a 64x or more increase in capacity per SSD >>die, without any large increase in cost or processing time. That translates directly into cost of >>product, and as 3D NAND hits its stride in 2015, we'll see 1TB SSD hit much lower price points,

You misunderstand the 3D roadmaps of the flash foundries.  The 32-48 layer chips now coming to market use much larger cells (~40nm) than the current planar flash. Stacking 32 cells that each take up 6-10 times the real estate doesn't give you 32x density, it gives you 4x density at best. Micron's new 3D chips have the same density per die as their 16nm planar. If you asumption of 32 layers being a 32x increase in density they would be much bigger.

Sure 3-D is the path to the future but it's how the flash foundries will continue to follow Moore's Law not how they'll break it and get 32x density in one step.

 >>Frankly, SSD doesn't even need to achieve parity to win. A 90K IOPS, 3Watt 1TB SSD at $150 will be very ??>>attractive compared with slow-as-molasses $39 1TB HDD in the server market, from archiving use cases all >>the way thorugh to tiered secondary storage to all-flash arrays. There won't even be a discussion in the >>virtualized server local instance storage case!

That might be true but again you made a very specific prediction and this point is irrelevant to whether your prediction comes true. Of course applications where IOPS are the measure of performance will go to flash.  No one is debating that. It's will it cost the same as an HDD like you predicted.

However there are many applications, most of which generate PB and PB of data, where sequential I/O is all that matters.  Streaming or survailance video, media archives, and the like are well served by big hard drives. One HDD can provide about 1/8-1/4 the sequential performance of an SSD if you have PB of video an object store with erasure coding will parallellize I/O to allow sequential I/O of several GB/s. Are you really going to store all the video from the thousands of police body cams on SSDs at even 4X the cost?

>>Even the capacity issue is a loser for hard drives. SanDisk has 16TB units planned for next year...that will be a strong competitor for that 6TB Seagate unit.

That depends on price, SSD vendors could have made 16TB 3.5" SSDs for a while.  There's very little demand because it will cost 5 times what the same capacity of spinning disk will and will bottleneck on the 6Gbps SATA connection. SanDisk has a 4TB SSD now, NewEgg sells it for $5,600. Density and price are not always related.

Then there's the business argument. Even if a foundry's costs fell by 10x per GB in 1 year in the short run there is no competitive pressure to drop prices faster. If vendor A has a 10x cost advantage but can still sell all the chips they can make at 30% cheaper they would be giving away money to sell for less than that 30% cheaper.  Until flash capacity increases dramatically in the 18 months of your prediction the market will be willing to pay prices higher than the 80% price cuts needed to make your prediction true.  Prices are driven by supply and demand with actual cost a secondary input.

My basic point still stands. Consumer SSDS cost 10 times what enterprise disks do per GB today. In 18 months that might be 5 times but it won't be even.

Re: NO Price parity before 2020

That's a long essay, Howard. I'll address just a couple of the major points.

First, your last statement. 

"Consumer SSDS cost 10 times what enterprise disks do per GB today". That's true...but they cost 20x just 8 months ago, and that was before 3D NAND. Even with the old rate of change, we'd see just 2.5:1 by the end of 2016, assuming the curve didn't change. The impact of 3D NAND will drive the improvement faster,

Moreover, as you've pointed out. SSD doesn't need to achieve parity to win. The performance benefits are so storng that users will pay a premium in many use cases. Also, with much lower power use, there is a TCO saving to be added to the equation. Power savings work out at around $8/drive-year, so add another $40 to the 5-year TCO balance and the hard-drive doesn't look so good.

The other point is that you seem to conflate interface protocol with drive class. That $600 drive was a so-called "enterprise HDD" with SAs interface. SAS "enterprise" drives are cheaper, but who needs SAS? The SATA SSD screamingly out-performs that SAS drive, even though SAS might get 5% more performance. Further, if the SSD is the same price as that SATA "enterprise" HDD, see the previous paragraph.

Re: NO Price parity before 2020

Jim,

You continue to obfuscate and dance around the question with "SSDs don't have to reach price parity" I can agree with that BUT you're the one that made the prediction. So are SSD prices going to fall 10X in the next 18 months as you predicted? That's what it would take for a consumer SSD to reach parity with TODAY's enterprise SATA HDD.

 

Even worse you continue to argue that the unsubstiatiated line about NetApp never having an SSD failure means that consumer SSDs are good enough for the data center. This is dangerous. People will follow your false reasoning and use cheap SSDs and therefore loose data.

Again let me say I know for a fact, from conversations with NetApp execs directly not an article I read somewhere on the internet, that NetApp does NOT us consumer SSDs but instead uses enterprise SSDs with much greater overprovisioning than consumer SSD use. This is engineering for greater endurance. They planned high endurance and they appear to have gotten it.

This data says nothing about consumer SSDs and to imply it does is a diservice to our readers.

 - Howard

Re: NO Price parity before 2020

Jim,

Actually, there isn't a 64x increase in NAND capcity per die on the horizon anytime soon. When they stack layers it doesn't mean that the density automatically increases as a multiple of the layers. For instance, Samsungs second-generation 32-layer NAND is only 86Gbit, which is less than most of todays planar (2D) NAND (which is 128Gbit). So, the capacity per die for Samsungs 2nd-gen V-NAND is less than that of todays planar. 

Intel's 3D NAND, which will not reach devices until next year, steps up to 256 Gbit per die. This is a mere doubling, and we arent even there yet. Each extra layer does not double capacity simply due to the fact that 3D NAND has a relaxed lithography (3Xnm, as opposed to current 16nm planar). 

The most optimistic projections (from Samsung and others) puts them at sampling 1 Tbit die (not in finished devices) in 2017.

Re: NO Price parity before 2020

I agree, Paul, which is why I phrased my comment on that as "By packing 32 or 64 times the capacity per die, 3D NAND will allow SSDs to increase capacity well beyond hard drive sizes."

We are at the start of a new journey with 3D NAND. As you know, the equation for drive capacity is very complicated, but the primary factor is die capacity, combined with die per wafer.

The first steps down the 3D path back off a couple of nodes (feature size reductions) to improve yield. The downside is that either capacity per die takes a hit, or the number of die per wafer goes down. The latter seems to be the path chosen today by most vendors. This offsets some of the cost reductions expected from the yield factor or total process steps reduction.

The Toshiba/SanDisk partnership has decided to stick with 15nm nodes, so they will trade yield for die size. They are also going for 48 layer solutions this year, which suggests they are seeing progress on the verticality issue. SanDisk has roadmapped 8TB drives this year and 16 next year.

It's also clear that vendors are being cautious and not pushing the limit of the number of cells per mm2. Even so, we can expect all of the factors to contribute to a dramatic reduction, exactly as the vendors have stated.

Moreover, there seems to be plenty of room for density evolution without changing node size. (This involves among other things being able to maintain vertical stacking through the layers better.) This growth potential contrasts with the hard drive which seems to have hit a technology wall leading to slow capacity growth.

I can't give interested readers URLs directly, but I suggest you Google "why 3D NAND doesn't yield 32x capacity" to get some sense of the issue.

Re: NO Price parity before 2020

Jim,

SanDisk is not using 15nm for its BiCS (3D) products. As a fellow member of the storage press I speak frequently with the company, so I can assure you the information is easy to verify. 

SanDisk's current 4TB SSD has mutliple PCBs (3) spread inside of a 2.5" 15mm thick form factor (Optimus MAX). As explained by Howard, this is very expensive. SanDisk is speaking of higher capacity SSDs soon, but these aren't going to be the standard 2.5" 7mm format SSDs. For 2.5" 7mm the SSDs are currently topping out at roughly 1.6 TB of user-adressable capacity (an Intel DC S3610).

There are no plans for 64x improvements in density, or even 32x, from any NAND vendor in the near future by any means of NAND manufacturing or packaging (such as simply stacking more dies). Increasing die density is the only way to achieve price parity, not simply packing in more packages per device (as is happening with the current 4TB versions, and will happen with the 8TB versions).

Re: NO Price parity before 2020

@Jim, well said "We are at the start of a new journey with 3D NAND".

It is difficult to imagine a consumer base that has UFS 2.0 memory at 600MB/s bandwidth (19,000 IPOS for random reading and 14,000 IOPS for random writing) uploading in the vicinity of 250 billion photos per day on Facebook, 100,000 tweets per minute to Twitter and 30 hours of video per minute to Youtube, etc., and datacenters around the world not reacting to changing market trends.

Cold storage comes into play but, cold storage is the secondary stage and performance is still the primary objective. 

Re: NO Price parity before 2020

Jim,

If you look at the history of NAND pricing, the $/GB decline has been more or less constant over the past few years. It's true that we saw a quick decline around 2010 timeframe, but that was due to overcapacity as the NAND vendors overpredicted the demand for NAND in the smartphone and tablet industry (everyone was betting that the NAND per device would increase as cost comes down, yet 16GB is still mainstream even in high-end smartphones). There is nothing in the horizon that would suggest a dramatic drop in NAND prices -- we will likely continue to see 15-20% price reductions per year similar to the recent years. 

3D NAND isn't a magical technology that suddenly brings NAND pricing down to HDD level. It's a vehicle for contuinuing NAND scaling, that's all. First generations of 3D NAND are actually less cost efficient than modern planar NAND nodes because the number of layers isn't high enough to mitigate the impact of larger lithography and increased number of process steps. That's why everyone is keeping their planar NAND around for at least the next few years, including Samsung who even introduced 16nm node after second generation V-NAND for more cost sensitive applications. 

Mark Webb has some good data on 3D NAND price modeling, so you might want to google that.

The short version is that 3D NAND likely won't achieve cost benefit over modern 15-16nm planar NAND until it reaches 64 layers, which is likely 2016-2018 depending on the vendor. Even after that point the cost will not drop dramatically because as with all semiconductor technologies one can't suddenly go from 64 to lets say 256 -- there are various technical obstacles that need to be overcome every time more layers are added. 

As for your "$99 250GB 3D V-NAND SSD" claim, there is nothing special about the price. Crucial introduced its MX100 over a year ago and the MSRP for the 256GB SKU was $110, which signals a 9% price cut in 12 months. There are some 256GB-class drives that regularly sell for $90 (e.g. OCZ ARC 100), so again the conclusion is that price decline is about 15-20% year over year. That's far from a "free fall".

SSD speed and durability?

The article quoted the great speed and durability of SSDs. I've seen issues with write speed due to how long it takes to erase a flash sector. Also, on durability, there are a limited number of erase/write cycles each cell can go through, especially if you are holding more than one bit per cell. I don't think magnetic storage suffers from these issues. However, I really like the idea of no moving parts! Is flash the ultimate technology for nonvolatile storage? How about FRAM? I don't know about cost on FRAM, but due to the popularity of flash, I suspect it's considerably higher. Remember bubble memory? I'm truly amazed with what HDD manufacturers have been able to do. I'm sure that eventually we'll move to solid state, but currently, rotating magnetic storage MAY have some advantages (write cycles, write speed, etc.).

 

Harold

Re: SSD speed and durability?

As I pointed out in a previous comment, the durability issue is mostly mythical these days, and better software masks the erase cycle question.

The bottom line is that these are the only questions against the technology, and they are essentially bogus!

Re: SSD speed and durability?

Sorry but the the stability of data in SSDs is a big question.

Unlike Hard Drives, SSDs lose their data at a much more rapid rate, particularly when the temperature is higher.

You cannot rely on SSDs for archival data.  Simply not.  They are too unreliable.

This puts a ceiling on how many SSDs can be sold.

And it insures that hard drives will always have a purpose - for backups and archives of data.

Re: SSD speed and durability?

I believe you are referring to the recent statement by Alvin Cox, an employee of HDD maker Seagate and a JEDEC sub-committee chair. This was covered by a quite a few news outlets, who viewed this as a doomsday scenario without doing any real analysis of the story. To their credit, Anandtech debunked the whole issue by simply pointing out that operating at 25C (measured on the drive) but storing at 55C is totally unrealistic. If anything the numbers are reversed in real life, due to the warming of the drive inside a server. Google "data retention SSD" and look at the Anandtech article.

The position that "SSD lose data" is baloney holds up in real life. SSD have been in mainstream use for 6 years now, and we've never seen a complaint of lost data. This myth is busted!

One value proposition of the SSD is that time to read data is much better than any other storage used in archiving. This suggests the archives will actually be powered up frequently, which might extend retention life indefinetly, especially if a patrol read is done occasionally.

Hard drives aren't trouble free in retention. Bearings can dry, air seals can leak without the positive pressure generated by spinning, and any contamination inside the drive can settle on disks or heads. One potential problem is that loose magnetic particles can end up on the recoring heads cuaisng a degaussing effect when the drive is spun up. 6-monthly spin ups make sense, and probably monthly is needed.

Tape, that old standby, is a bit of a disaster long term. Print-thru can occur, and tape layers can stick to each other. Serious long-term storage of tape needs 6-monthly reading of the whole tape end-to-end to prevent this.

All-in-all, it sounds like ANY archiving approach can't be totally passive. ALL of the approaches need the power to be turned on regularly. If that's done, SSD will be a perfectly good archiving approach once the price drops kick in.

Re: SSD speed and durability?

"And it insures that hard drives will always have a purpose - for backups and archives of data"

If you seriously think that hard drives play an important role in 20 years, then you're seriously fooling yourself. You also seem to forget that mechanical drives are really reaching their limits.

Re: SSD speed and durability?

Thanks for the comment! How many erase/write cycles can a flash cell survive? A flash device I use a lot (M25P16) is rated at 100,000 erase/program cycles. I do not believe magnetic drives have any such limitation. In most use cases, however, we may never write a sector 100,000 times, so this would not be an issue. It all depends on how the drive is being used. But, if there is indeed a limit like this, I don't believe durability is a myth. On write speed (again, looking at the chip I'm most familiar with), a sector erase time is up to 3 seconds with a typical time of 800ms. You can, of course, buffer write data during this time, but you does seem that you hit a speed limit due to sector erase time.

Harold

Re: SSD speed and durability?

Of course durability is an issue. It is in all computer gear. The myth is that SSd will fail overnight - that just isn't rue in real life.

Carnegie-Mellon and Facebook just published an exhaustive study on failure rates in the Facebook population of PCIe Flash cards that describes a failure model for those devices. It's a good pioneering effort to understand the mechanisms at work and to improve the product, but there are already signs that later generations are doing better than older product.

It was interesting that the next article I read said "Apple recalling hard drives from 2012 and 2013". This is a common event that didn't figure in the hard-drive bathtub curve model. I suspect, as do many, that the floods in Thailand had a wider impact than we thought, and that quality stndards may have lapsed for a while after the restart.

Effects like these require a more subtle approach to archiving than backing up the data and tossing it in a drawer. Multiple replicas are needed, with strong error and failure correction. This applies to tape too, though it's actually harder to do there..

Don't sell tape short

While SSDs always seemed the way of the future, simply by the fact that they don't have any moving parts and are a lot more durable therefore, I wouldn't write off tape just yet. IBM's recent progressions with the technology make it much cheaper and offer huge storage capacities. 

Yes, a lot of data needs to be quickly accessed, but most of it doesn't. Audio and video recordings for secuirty purposes, accountability records and historic data that may one day useful but for now is redundant, all of that long term data needs to go somewhere. Tape is perfect for that. 

Re: Don't sell tape short

"Yes, a lot of data needs to be quickly accessed, but most of it doesn't. Audio and video recordings for secuirty purposes, accountability records and historic data that may one day useful but for now is redundant, all of that long term data needs to go somewhere. Tape is perfect for that. '

Whoopty, I don't think there is any drastic change in read and write speed. But to an extent, it can cater the growing storage requirement and place to hold it (in terms of size).

Re: Don't sell tape short

Is tape the best solution?

The photo approach actually doesn't fit modern usage too well. Facebook figures that most of its photos over 12 month old are archivable, but how do you explain to a user that you have to wait an hour to retrieve that 5 year old photo? Ideally, the archive is cheap and low-power, but able to turn on and deliver data fast....Tape doesn't fit the bill.

The issue of cost is important today, too. LTO cartridgesw aren't much chepaer than hard drives, measured on a dollar/terabyte basis. When you amortize the cost of a petabyte tape library across that, disk is cheaper! Demounting tapes and sending them to the salt mine costs money, so the saving for tape is marginal.

Re: Don't sell tape short

"The photo approach actually doesn't fit modern usage too well. Facebook figures that most of its photos over 12 month old are archivable, but how do you explain to a user that you have to wait an hour to retrieve that 5 year old photo? Ideally, the archive is cheap and low-power, but able to turn on and deliver data fast....Tape doesn't fit the bill.'

Jim, you are very right. Archiving is very easy, but retrieving it when required is a time consuming process. In such situation, whether you have any solutions to suggest.

Re: Don't sell tape short

Another alternative is to utilize compression technologies. For instance, IBM's Informix can compress data up to 80% and it would save cost to compress data before it is stored. Compression creates a heavy workload on computation. Nevertheless, computational power has been exponentially increasing as compared to storage technologies and hence, it will not create a problem.

Re: Don't sell tape short

"Another alternative is to utilize compression technologies. For instance, IBM's Informix can compress data up to 80% and it would save cost to compress data before it is stored. Compression creates a heavy workload on computation. Nevertheless, computational power has been exponentially increasing as compared to storage technologies and hence, it will not create a problem."

Brian, that's a good idea but compression and decompression, is time consuming process.  Actually we need some technology for storing more datas per segment. 

Less space more data

"By packing 32 or 64 times the capacity per die, 3D NAND will allow SSDs to increase capacity well beyond hard drive sizes. SanDisk, for example, plans 8 TB drives this year, and 16 TB drives in 2016. At the same time, vendors across the flash industry are able to back off two process node levels and obtain excellent die yields."

Jim, this is amazing!! I think the main intention is more storage at less space. Am eager to know whether the data representation technology is changing or still deal with old way of writing bits per sector.

Well written article Jim

Well written article Jim, I pretty much agree with everything you wrote. Even hard drive manufactuers will silently agree. Any hard drive manufacturer who thinks differently will be doomed. In 10-15 years hard drives will not pay an important role anymore. Hard drive manufacturer need to be prepared for this now and not in 10 years.

Sorry, prediction is too bold

I've watch HD come down in price over a steady curve for 30 years. During that time, there were minor blips from overinventory or shortages from floods, but a steady line for the most part, no step fuction of 50% drops overnight. SSD has followed a nice curve as well, and right now we are at $35/TB for HD and just over $225/TB for SSD, or a ratio of about 7X. It would be bold to predict this ratio would drop below 5X by the end of 2016, but that would be the best I'm expecting. At that point, $20/TB for HD and $100/TB for SSD. 

I'd love to see an 8TB SSD for $200. And I will, just not for 5-7 years.