Earlier this month Western Digital’s HGST division invited The SSD Guy to a launch of a number of products. On the HDD side there were:
- 6TB air HDD, HGST’s last air-filled enterprise HDD
- 8TB helium HDD, an incremental upgrade of last year’s 6TB helium HDD
- 10TB shingled helium HDD (pictured)
I view these as very solid evidence that HDD costs will continue to stay an order of magnitude cheaper than SSD costs, thwarting the price-per-gigabyte crossover that others have been predicting for years.
In fact, since my last post on the price crossover in 2011, very little has changed.
It’s safe to assume that the HDD industry will follow HGST’s lead with competitors shipping both enterprise and client versions of these capacities. As HDD makers always do, they will then drive the price of these products to the same level as their predecessors, or roughly $50 for the client HDD & $350 for enterprise version. This will allow the price per gigabyte for HDDs to drop significantly over the next few years.
HDD capacities typically increase in lurches, with vendors extending one technology as long as they can, then moving to the next to cause price per gigabyte to drop rapidly for a while. This happened in the past with the onset of air bearings, giant magnetoresistance (GMR), perpendicular recording, and other technologies that The SSD Guy, because of his semiconductor focus, doesn’t know much about. We are now entering another phase of rapid price drops enabled by the move to helium and shingled recording.
This change will once again prevent SSD prices from approaching the price of an HDD.
Assuming that a client version of HGST’s 10TB shingled HDD sells for $50 three years from now the price of HDD storage will drop to $5/TB. If NAND successfully follows its historical price decline of -40%/year (starting in late 2012 when prices stabilized) it should cost $30/TB by then.
Most people think that a 40%/year NAND price decline is no longer attainable. If NAND prices decline at the less aggressive 30%/year price reduction that DRAM has historically followed then NAND would cost $67/TB by mid-2017.
NAND prices would have to drop by an average of 58%/year to reach price parity with a $5/TB HDD by mid 2017. It’s pretty certain that this isn’t going to happen!
At Objective Analysis our business is to help our clients understand how the future will impact them. When our clients ask about an SSD/HDD price per gigabyte crossover we provide this sort of analysis to show that HDDs will enjoy an ongoing price advantage over SSDs. Please contact us to learn how you can take advantage of our insight and understanding to gain a competitive edge.
For each application there is a maximum useful capacity. Most consumer electronics that need up to ~100GB have already crossed over. An interesting analysis would be the cross-over capacity, I believe that today it is between 100GB and 200GB. SSDs will slowly gobble up consumer products, including the PC. HDs will become higher capacity and more expensive over time, but we may not see them disappear completely for many years. After all, if capacity was the most important parameter, we would still have 8″ drives.
Good observation Mr. Kagan.
I am glad that you said what today’s useful capacity is, since we will certainly look back at that number in ten years and find it amusing.
Your comment about 8-inch HDDs is very well put!
For a number of years I have argued that $50 is the lowest price for an HDD, so any amount of flash that costs less than $50 will drive designers to use flash rather than an HDD. Today, at about 40 cents per gigabyte, that would be about 125GB. This seems to match your thinking.
Jim
Comparing SSD and bulk storage drives is a bit like comparing a Ferrari with a semi truck. They do very different things.
If you compare SSD and ENTERPRISE HDD, it’s clear that SSD are often cheaper, and always much faster.
Yes, Jim O’Reilly, you’re right.
What I was addressing were the folks who say that the Ferrari will eventually grow to have the hauling capacity of the semi truck without giving up any of its speed or handling.
Jim
Actually, it’s not true that SSDs are “always” faster. In sequential I/O patterns, HDDs can stream data just as fast, if not faster, than an SSD for the same $. So, if you could intelligently write data to HDDs in a manner that takes advantage of that technology’s strengths, such as a log structured file system which leverages CPU cores for performance, then you can get flash-like performance or better from HDDs. It’s about reducing the random I/O burden on disk. Of course leveraging SSD as a random read cache, at the very least, is still a good idea.
Yes, Brandon, you’re correct, and some designers use this fact to make their disk systems faster than their competitors’.
I even wrote a blog post about it right here on The SSD Guy some time back: https://thessdguy.com/an-hdd-cache-for-an-ssd/
In that system some University of Toronto researchers used an HDD to help their SSDs run faster!
Most applications access the drive randomly, though, so, as a general rule, Jim O’Reilly is correct.
I am curious to know then what you think will happen when the data path moves from SAS channels to PCIe NVMe?
If you leverage the new data paths when SSDs get denser (Samsung have announced a 3.2TB NVMe based SSD), where NVMe can help reduce latency even more in an SSD, will that not allow the Ferarri to act as a Semi?
Didn’t a 14TB SSD also get announced at the Flash Memory summit for release in 2016. Surely that’s got to start to impact on SATA HDDs?
Alex,Thanks for the comment.
Our survey respondents have explained to us that SAS SSDs will stick around for quite a while to support existing SAS-based systems. PCIe/NVMe shipments already exceed SAS unit shipments and will remain in the lead through our forecast window. All this is in the report on Enterprise SSDs available on the Objective Analysis website.
The point of this post, though, wasn’t about interfaces, but about the fact that HDDs will continue to have a price per terabyte that is an order of magnitude lower than that of flash, and that predictions of a price crossover are ill-founded.
Jim