SSD price premium over disk faaaalling The general solid state drive price premium over disk should decline from 6.6x now to 2.2x in 2021, according to new IDC numbers. IDC updated its SSD forecast last week and Stifel analyst and MD Aaron Rakers has had a look at the updated forecast. It says SSD shipments will increase from around 157.5 million units shipped in …
Why wouldn't hard drives completely disappear, once the price premium disappears? OK sure, SSDs will lose information if left unpowered in a safe for a few years while a hard drive is probably OK, but tape is probably a better alternative for such situations. Yes, I'm saying tape will outlive hard drives!
The premium won't even have to go to 1x for hard drives to go away. The more it goes down, the smaller the market for hard drives becomes. Once cloud providers feel SSDs are better than hard drives even for cold data and even if they still cost a bit more (because they are smaller, no spinup delay, use less power, or whatever) and they stop buying, that's pretty much the end of the market. Even if a few niche cases like 'keeping hard drives in a safe' remain, sales volumes will be too low to make it worthwhile to continue production.
"Why wouldn't hard drives completely disappear, once the price premium disappears?"
First, the premium will never entirely disappear, simply because the fixed capital requirements for SSD production are orders of magnitude higher than HDD and storage requirements grow faster than SSD production can. Even at current price multiples, it takes 4-5 times as long for an SSD fab to pay for itself compared to an HDD plant, and it's probably not possible for the premium to sink below 2x ever, unless someone finds a NAND tree.
Second, there's some circumstances where raw capacity is always preferable to speed, so as long as the HDD is at a lower $/TB, it'll be the better choice. Consider, say, FB photo storage, which is rarely accessed but needs vast capacity and increases enormously and continuously. You just want to stick in the largest capacity item possible at the lowest price possible. That's going to be HDD.
There probably is a technology coming in the near future which will obsolete HDD, but it's not going to be SSD; we need something that achieves higher speeds at HDD-equivalent prices. Either that, or some breed of NVRAM will come along and simply eliminate the concept of separate storage.
I guess you don't understand economies of scale, if you could think that fixed capital requirements mean the product with a larger up front capital investment can never be cheaper. If you have fixed capital requirements of $1 million and variable costs of $50 versus fixed capital requirements of $1 billion and variable costs of $10, then if you have enough volume the $1 billion fixed capital requirement products will be FAR cheaper.
The only variable costs for making flash chips are the per wafer processing costs, which are less than a thousand dollars per wafer - regardless of the number of bits the chips you carve from that wafer hold! While the fixed capital investment for a new process that doubles the number of bits per wafer is huge, it is nowhere near double the capital investment for the previous generation process, thus the cost per bit is constantly being driven down - and at a much faster rate than the cost per bit of hard drives is being driven down!
How else do you think that SSDs have gone from 100x premium to 20x premium to 5x premium over the years? If some new breakthrough that let hard drives double in storage every couple years (like was happening in the early 2000s thanks to the GMR technology IBM developed) then they'd be able to keep pace with flash, but that's very unlikely to happen since R&D into hard drives is a fraction of what it used to be. Otherwise unless we stop being able to put more bits on a NAND chip, SSDs will catch up with and then be cheaper than hard drives.
The fixed capital requirements will be higher for spinning disk simply because you have less revenue to justify it. If the market for spinning drives reduces too much, there is not enough of a market to justify the expense of building a factory.
Your argument that raw capacity/TB is sometimes completely preferable to speed is possible but I doubt it. If the cost includes the lower power for such endeavors and much smaller rack space, the difference in $/TB would be factored out quickly as the SSD sizes ramp up. Not to mention you can already get a bigger SSD than any spinning media right now. How does the argument work when spinning drives are much, much smaller? The manufacturing costs are being driven down with SSD and very little R&D is going to spinning rust. The latest shingled tech involves relatively little capacity gain and terrible reliability. I may be missing something but from what I see, spinning drives are going the way of the vacuum tube. It's OK. They really were great engineering but it seems silly to have spinning disks when you can have solid state and not worry about vibrations, vacuums, helium, and such.
"If the market for spinning drives reduces too much, there is not enough of a market to justify the expense of building a factory."
That's just circular logic. You're saying the market for spinning rust will shrink because it's out-competed by flash, and it will be out-competed by Flash because the market for spinning rust will have shrunk.
As for refuting DougS's points, I'll leave that to the sage words of, um, DougS in this thread, from when he argued the exact opposite: https://forums.theregister.co.uk/forum/1/2014/12/09/no_flash_datacentre_takeover/
If you read that post of mine you dug up from 2 1/2 years ago, you'd see I was replying to the idea that hard drives and SSDs would reach price parity in two years (i.e. now) and I suggested it would be around a decade before there would be enough NAND capacity in the world to satisfy all storage demand.
Which, surprise surprise, is about the 2024/2025 timeline we're talking about for price parity and hard drives going away...
"Why wouldn't hard drives completely disappear, once the price premium disappears? OK sure, SSDs will lose information if left unpowered in a safe for a few years while a hard drive is probably OK, but tape is probably a better alternative for such situations. Yes, I'm saying tape will outlive hard drives!"
Tape has its own problems with longevity. You may have trouble finding a compatible machine that can read it.
http://spectrum.ieee.org/computing/it/the-lost-picture-show-hollywood-archivists-cant-outpace-obsolescence
The same would be true with hard drives, if you had an IDE hard drive you'd need something to read it. If you have thousands of DLT tapes, you should probably keep a few DLT drives in the archive as well (and hardware to read them - same problem as if you had IDE drives...)
No physical storage hardware is immune from this. Cloud has an advantage here, though you'd need to store to multiple clouds to insure your information was recoverable in case e.g. Google Drive disappears at some point like Google stuff seems to do sometimes.
are really only interested in getting rid of our own stack of rotating rust.
At the moment, the really sweet spot seems to be 512Gb SSD's (PC SATA types).
Anything over 1TB still costs an arm and a leg (or two)
Until I can get a 4TB SSD for the same price as I now pay for a 4TB block of rotating rust then articles like this really mean diddly squat.
That is the real world not how someone paid to spek out of their 'anal' thinks it is.
Your needs are not everyone else's needs. For myself, I realized that neither my desktop nor my laptop were actually using the 1 TB drives they had and downsized trading the wasted space a very noticable speed increase. Same goes for things like my home firewall, it didn't need anything more than the 64 gb drive I gave it. I still keep two spinning drives in my home NAS (video collection is large) but I have more SSD drives than spinning in my home.
At work, it's pretty much the same story, I have a lot of wasted space on my servers and when they get upgraded, I will downsize the space and trade in for SSD.
Yep, I bought a 250 GB SSD to replace the 500 GB hard drive my laptop came with. I would have bought it without a hard drive since I knew I was never going to use the drive it came with, but that's unfortunately not an option. I wonder how many hard drives are sold in laptops these days that are immediately removed and put on a shelf to gather dust and never be used?
"I wonder how many hard drives are sold in laptops these days that are immediately removed and put on a shelf to gather dust and never be used?"There's room enough for two snapshots of the SSD on the spare and the HDDs are small enough to go in the shoulder bag.
Until I can get a 4TB SSD for the same price as I now pay for a 4TB block of rotating rust then articles like this really mean diddly squat.
And the bloody price of rust has stopped nose diving, I could do with swapping out the 2TB drives in my fileservers at home for 4TB ones, but when you need to buy them 4 at a time they're too pricey.
Take a look at how much your swap file is actually used. My remaining Pi version 1 has 1GB of swap space, 252K of which is currently in use. It still uses its original 8GB SDHC card. The 7th field of /sys/class/block/mmcblk0p3/stat divided by uptime gives 52 sectors/day. SSD's are guarantied for years with the assumption of one complete drive write per day. 3½" disks are guarantied for years with the assumption of one spin up / spin down cycle per day. (If you want more cycles, pick a 2½" drive.)
That Pi started with one spinning disk, and currently has three. The oldest disk got replaced a few months ago when it started to make unusual noises. Pi V1 has only ½GB of ram. Just about everything under five years old has much more, and even less use for swap. If I had something writing huge amounts of data I would still prefer SSD for reliability, and if you are going over 6 drive writes per day, a spinning disk is probably too slow to read the data back even once per day.
Take the write lifetime of a quality consumer SSD and divide by 100 MB/sec (about the fastest sustained write you can get from a single hard drive) and you can still count on a couple years of service, more if you make that your main buying criteria. Obviously if you are pushing data as fast as the SSD can handle that lifetime will be much shorter, but in that case you'd have to compare with an array of striped hard drives to deliver the same throughput!
If you compare a solution designed for the 600 MB/sec SATA3 limit you'd need to compare a half dozen hard drives striped together (plus a SATA controller card to handle them, and external box to put them in) versus a single SSD. Even if you spend more on that SSD to get one with a higher DWPD (using one with more capacity than you need is a good way to do that, or buy an enterprise model) the cost crossover is a lot sooner for this than it is for a single drive versus single SSD...
Data churn isnt that bad. My desktop has an old 250gb OCZ agility. It is about 4 years old daily use. It hasnt started remapping yet.
My laptop is a 3yo heavily used samsung 250gb and that says 95pct life left with 0 remaps. That is an MLC too.
An email server will probably fare worse of course.
I think a lot of installed HDD capacity is going unused, especially in domestic PCs and that will tend to increase as areal densities increase. Manufacturers of PCs know that if they have a choice of fitting a 1TB HDD or a 500GB SSD at the same price, the 2x price premium of SSDs is not sufficient to choose a HDD.
HDD manufacturers could offer crippled drives at a lower price, but that won't change the manufacturing cost. One option might be to shift to making a lot of drives smaller than 2.5" .
They could offer a 160 GB SSD or 1 TB hard drive at the same price, and let the buyer choose. A lot of people will be stupid and choose the hard drive, but at least those who are being helped by people who know what they're doing will get them to make the right choice - because 160 GB is probably more than enough for 99% of people who don't understand that choice.
Storage capacity on consumer desktops goes underused in general. I would not expect a switch to SSD to alter this.
The bigger drive will always be chosen by the consumer PC vendor. They want to sell a bigger number. The cheapest way to achieve that will be chosen. Because the consumer market is all about pushing the cheapest piece of crap they can manage.
The only people that would care to pick the smaller SSD are also those more likely to need more capacity than SSD can handle.
Doubt they are particularly counting on new technology. The two factors that are probably most responsible are new entrants/factories increasing supply allowing competition to drive down prices, and over 5 years the upfront capital investment should have been written off (or down at least) reducing the production costs per unit.
"The chart shows the mission-critical enterprise SSD premium rising from 15.9x in 2016 to 17.1x this year and falling to 14.2x next year when, presumably, supply constraints ease."
No, the chart shows the mission-critical enterprise SSD premium falling from 3.4x in 2016 to 3.2x this year, and 2.4x next year.
A couple of years ago I bought a huge monstrosity of a laptop so that I could have 2.5TB storage space. That included standard "large" form factor drives both SSD & HDD.
Now a mere 2 years later for the same price I can buy from the same vendor a similarly equipped machine that looks more like a MacBook than an 80s luggable. It's all smaller form factor SSD.
RAM and SSD prices have gone up because of a global shortage caused by demand outstripping supply. Everyone and their dog are currently building new factories which should mean that by the end of the year supply is again greater than demand and prices will fall after that.
Of course, if you're in the UK prices have also gone up because of the Pound's collapse against the Dollar.
Tell me about it. I built a 5820k based rig last year. I bought all new kit apart from the two main 250GB SSDs that I had already.
This time last year it all came to £850.
The same kit today would cost £1000!
I've stopped buying SSDs to upgrade customers machines where possible. I was buying 120GB SSDs for 30-£35 a pop last year but now we are in the £55+ range.
Sounds about right. I would have expected the gap to be even less by 2021.
What I don't understand are the vendors who are arguing that SSD and nearline have the same or appx the same price point. For instance, IBM isn't making XIV Gen4 because, as they argue, why would they make a, primarily, nearline SAS array when they can provide SSD at the same price point. I don't see how that is possible. Same price point per I/O makes sense, but not sure how they would provide the same price point for capacity.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
