[reader50]

I was looking at my newegg order history, and realized how badly we're being ripped for hard drives today.

In 2010, I bought a couple 2TB Samsung drives, 7200 RPM. $80 each on sale ($40 per TB). I think they were usually $90. Today, the best prices I can find are:

2TB newegg: $68 (generic, 7200 RPM) [$34/TB]
3TB newegg: $90 (generic, 5400 RPM) [$30/TB]
3TB newegg: $100 (generic, 7200 RPM) [$33/TB]
4TB newegg: $150 (generic, 7200 RPM) [$37.50/TB]
5TB newegg: $230 (seagate, 5900 RPM) [$46/TB]
5TB amazon: $175 (toshiba external, 7200 RPM) [$35/TB]
6TB newegg: $280 (WD green, 5400 RPM) [$47/TB]

Based on Moore's Law, drives should have dropped to $20 per TB over the last 4 years. Or even less. 4TB for $80-90, 5TB for $100-120, 6TB for $130-150.

Instead we got market consolidation and slow price cuts. Yay. SSDs have dropped in price during this time, from 30x HD cost to 10x HD cost. They're approaching $400 per TB. The ratio has improved mainly because HD prices barely budged.

I'd been waiting for HD prices to correct, and could not wait any longer. It's not fun to be robbed.

[BLAZE_MkIV]

Moore's Law applies to transistors so you could apply it to SSD drives not magnetic media drives.

I was told that SSDs have a limited number of times that one can read/write data to it before you can't use it anymore.

Is this true?

If so, then maybe manufacturers should extend the limit so these drives last longer, before improving their capacity.

(I still use my old 2.5" 120 GB disc rescued from a laptop as my time machine backup. It rattled in a case and still hasn't failed. It will someday I'm sure...) But is it true that SSD has a limited lifetime based on the number of reads/writes?

[P]

Yes, sort of - a cell can be written to some 3000 times on newer drives. It's actually worse than that - a cell must be deleted before it can be written to again, and the deletion process must happen on a much bigger segment than one bit. This means that to even get to 3000, you have to clever about how you write to the cells. And no, they're not working on allowing more writes. It is actually going the other way - cells used to last for 10 000 cycles.

The reason all of this works out is wear leveling. By making sure that you write to all cells equally, you can make sure that drives last longer. Effectively a 250 GB drive can support 250 GB times 3000 cycles equals 750 TB. That takes more than a decade for a regular user to write, and a spinny disk would have died from mechanical failure long before that. In fact SSD failure rates are generally much better than HDD rates, and the failures much more obvious.

There is some hope for reliability, though. Samsung has recently launched the 850 Pro, based on so called V-NAND, which brings write reliability back towards that 10 000 mark while simultaneously boosting potential capacity. The 850 Pro is still an expensive enterprise drive, but the regular 850 is supposedly coming.

Oh, and you can read from cells as many times as you like, and it keeps charge for at least a year after the last write.

[P]

Originally Posted by reader50 

I'd been waiting for HD prices to correct, and could not wait any longer. It's not fun to be robbed.

HDD manufacturing is a cartel. They increase prices when they feel like it. They've also been hurt by the rise of flash in the enterprise, as that was their most profitable segment.

[reader50]

Originally Posted by BLAZE_MkIV 

Moore's Law applies to transistors so you could apply it to SSD drives not magnetic media drives.

Platter areal density is equivalent, and it has more than doubled in the last 4 years. Most of the HD cost is the mechanicals, which creates a lower price limit of ~$50 for 3.5" drives. Manufacturing cost goes up a little with higher platter counts. So smaller drives in a generation tend to cost $60-80 regardless of size, while bigger drives in a generation likely cost less than $100 to manufacture.

Originally Posted by P 

Oh, and you can read from cells as many times as you like, and it keeps charge for at least a year after the last write.

Do you have any links for SSD archival persistence? I've had trouble finding hard data.

[akent35]

Originally Posted by P 

There is some hope for reliability, though. Samsung has recently launched the 850 Pro, based on so called V-NAND, which brings write reliability back towards that 10 000 mark while simultaneously boosting potential capacity. The 850 Pro is still an expensive enterprise drive, but the regular 850 is supposedly coming.

Oh, and you can read from cells as many times as you like, and it keeps charge for at least a year after the last write.

Great info, P! Do you know if a Samsung 256 gig 840 Pro (I put one of those inside my Mac Mini last November) is similarly more "reliable" than a Samsung 256 gig 840 EVO drive? (Although the price difference is getting smaller, the 840 Pro is still more expensive than the 840 EVO).

[Mike Wuerthele]

Originally Posted by akent35 

Great info, P! Do you know if a Samsung 256 gig 840 Pro (I put one of those inside my Mac Mini last November) is similarly more "reliable" than a Samsung 256 gig 840 EVO drive? (Although the price difference is getting smaller, the 840 Pro is still more expensive than the 840 EVO).

Too early to tell. The whole MLC, TLC, SLC reliability thing still has yet to see enough mileage in the real world for super-precise determination of lifespan.

That said the 850pro (we're testing two of them) is warrantied for 10 years.

[akent35]

Originally Posted by EstaNightshift 

Too early to tell. The whole MLC, TLC, SLC reliability thing still has yet to see enough mileage in the real world for super-precise determination of lifespan.

That said the 850pro (we're testing two of them) is warrantied for 10 years.

Thanks, EstaNightshift. Here are some of the specifications of the two drives I mentioned:

Samsung 840 EVO (MZ-7TE250BW):

Sustained Sequential Read
540MB/s

Sustained Sequential Write
520MBps

4KB Random Read
Up to 97,000 IOPS

4KB Random Write
Up to 66,000 IOPS

MTBF
1,500,000 hours

Samsung 840 Pro (MZ-7PD256BW):

Sustained Sequential Read
540 MB/s

Sustained Sequential Write
520MBps

4KB Random Read
Up to 100,000 IOPS

4KB Random Write
Up to 90,000 IOPS

MTBF
1,500,000 hours

So, at least from these figures, the primary differences are:

1. The 840 Pro is 6 gig larger.

2. Both the 4KB Random Read and 4KB Random Right figures are larger on the 840 Pro (very little difference between the 4KB Random Read figures, but about a 36% difference on the 4KB Random Write figures).

[P]

The difference between 840 Pro and 840 Evo is the type of flash used. 840 Evo uses the quite new TLC NAND, which is theoretically less reliable than the old MLC NAND in 840 Pro, but as Esta says, we do not have any data yet with TLC being so new.

The vertical NAND used in the 850 Pro increases capacity by a factor of 32 by stacking the cells vertically. This improvement is big enough that Samsung has elected to go back to an older process node, which increases the number of writes you can safely do. If they had made old flat SLC or MLC flash on that node, the rating would have been 10 000 cycles (as opposed to about 3000 for the newer nodes). With V-NAND being so new, Sammy has elected to rate it at 6000 cycles for now. I guess they figured double the performance was a pretty good improvement.

(At the same time, Samsung's marketing is talking up "10 times better reliability", which would imply 30 000 cycles, but then their marketing department is not known for being overly concerned with the truth. They could be comparing to some bargain-basement crap SSD with terrible wear leveling).

[P]

Originally Posted by reader50 

Do you have any links for SSD archival persistence? I've had trouble finding hard data.

That statement (read for a year after the write cycles are out) comes from here:

AnandTech | Testing Samsung 850 Pro Endurance & Measuring V-NAND Die Size

which in turn gets it from the JEDEC specs, which should be reliable. I do not have any better info on how long you can usually read a drive while it still has write cycles left.

[Waragainstsleep]

The HDD prices got too low a few years back. There was a fire in a factory back then when the 2TB drives bottomed out for price and the prices spiked as demand outstripped supply. They have barely gotten back down to those levels from 2010. It doesn't help that there are now only two players in the HDD market.
On the plus side Seagate are about to launch an 8TB 3.5" unit so perhaps that will help a little.

I think you also need to look at the consumer market. 2TB is still plenty for the majority of users. Most people still never fill a 500GB drive on their Windows box, those who stash a lot more media on their rigs are probably still struggling to fill a 2 or 3TB before it fails from old age. If more people were deciding they needed 4TB drives in order to keep a months worth of 4K viewing or something, maybe the prices would start to go down again.

[reader50]

Originally Posted by P 

... and it keeps charge for at least a year after the last write.

Originally Posted by reader50 

Do you have any links for SSD archival persistence? I've had trouble finding hard data.

Originally Posted by P 

That statement (read for a year after the write cycles are out) comes from here: ...

I meant, how long does the non-volatile SSD memory last without power? Unplug an SSD, set on shelf. How long can it sit before the data becomes uncertain? Your first quote suggests at least a year, and wikipedia also mentions in passing that SSDs can't hold data indefinitely without refresh power. But I haven't seen any hard data.

[Mike Wuerthele]

Originally Posted by reader50 

I meant, how long does the non-volatile SSD memory last without power? Unplug an SSD, set on shelf. How long can it sit before the data becomes uncertain? Your first quote suggests at least a year, and wikipedia also mentions in passing that SSDs can't hold data indefinitely without refresh power. But I haven't seen any hard data.

In THEORY, the data could last permanently. In REALITY, that's not so likely. In floppy days, I called it bit rot, but things like cosmic rays, harsh magnetic fields, all have impacts on stored data in flash form.

I don't think anybody's done any real work on it.

So why are manufacturers moving away from SLC even though it gets 10,000 cycles? Can't we just go back to the SLC?

[Mike Wuerthele]

Originally Posted by And.reg 

So why are manufacturers moving away from SLC even though it gets 10,000 cycles? Can't we just go back to the SLC?

MLC and TLC have a higher data density, and are cheaper per unit GB. They're the major reason for the drop in SSD prices.

[P]

Originally Posted by And.reg 

So why are manufacturers moving away from SLC even though it gets 10,000 cycles? Can't we just go back to the SLC?

SLC means that each cell can store 1 bit, MLC stores 2, and TLC stores 3. Since SSD $/GB has been so high, getting capacity up has been the top priority, and since drives with good wear levelling do not run out of write cycles in real world situations, that was less of a concern.

When I say that the number of cycles has dropped from 10,000 down towards 3000, I don't speak about SLC versus MLC, though - I speak about the process used. In semiconductor manufacturing, moving to a smaller node is good, because you can fit more transistors on the same area, so if you go from (say) 65nm to 45nm, suddenly you get twice as much storage for the same die area (and theoretically the same price)*. The problem for flash is that if you do that, you also lose write cycles (or P/E cycles as they're usually called), so with flash now being produced on things like a 15nm process, the number of cycles has dropped quite a bit. The advantage with V-NAND is that the capacity goes up so much that you can hold off on using the newest processes. Samsung went back to 40nm, I think, which means that the write cycle count went back up to 40nm levels.

* Same price if the price per mm2 is constant. In the last few generations, it hasn't been, because investments have gone through the roof, and going from 28nm logic to 20nm logic apparently yields no cost advantage at all. Also the defect density goes up, at least initially, and must be compensated for.

[P]

Originally Posted by reader50 

I meant, how long does the non-volatile SSD memory last without power? Unplug an SSD, set on shelf. How long can it sit before the data becomes uncertain? Your first quote suggests at least a year, and wikipedia also mentions in passing that SSDs can't hold data indefinitely without refresh power. But I haven't seen any hard data.

I know what you mean, but I don't know the answer. We have two data points:

* The JEDEC spec that a drive with exhausted write cycles should be readable for one year after that
* The ten year warranty that some enterprise drives have.

[Spheric Harlot]

Waragainstsleep:

HD prices weren't driven up by a fire. The world permanently lost something like a quarter of its hard drive manufacturing in the floods in Thailand in 2011, and much more of that in the short term (like, a year or two).
Prices quintupled over a few weeks and stayed there for months until factories slowly started getting restored, and the development really kickstarted adoption of SSDs as the price delta became smaller.

In the (literal) wake of those floods, several manufacturers fused, and with the obvious rise of the SSD, there was little incentive to roll back to previous manufacturing capacity.

So even after recovery, prices for hard drives were thrown back by about three years and are working from there

[noibs]

I believe there are two answers to this question and both have been posted above.

First are the floods in the Pacific Rim that took out a large part of the hard drive manufacturing capacity.

Second is the consolidation of the hard drive industry. There simply isn't the competition that there once was. Regardless of brands, aren't there now only two main manufacturers (or maybe three).

[akent35]

Originally Posted by noibs 

I believe there are two answers to this question and both have been posted above.

First are the floods in the Pacific Rim that took out a large part of the hard drive manufacturing capacity.

Second is the consolidation of the hard drive industry. There simply isn't the competition that there once was. Regardless of brands, aren't there now only two main manufacturers (or maybe three).

Good points! As far as remaining prominent manufacturers of hard drives (non-SSD), I can think of at least two of them: Seagate, and Western Digital. Any other ones?

[reader50]

Toshiba is the 3rd. Euro regulators approved WD buying HGST, but required enough assets be sold off to Toshiba so they could immediately enter the 3.5" consumer market.

Toshiba is around, but isn't competing much in the 4+ TB markets. Yet.

[dxtr]

Originally Posted by reader50 

Toshiba is the 3rd. Euro regulators approved WD buying HGST, but required enough assets be sold off to Toshiba so they could immediately enter the 3.5" consumer market.

Toshiba is around, but isn't competing much in the 4+ TB markets. Yet.

You are correct there are now only three.
Toshiba is less expensive but in practice much slower that either WD or Seagate.

[ghporter]

Industry consolidation is hard on consumers, but there is still the rather heavy load on consumer prices caused by the floods 3 years ago; vendors, distributors, and yes manufacturers, liked the higher prices, and they are really only just now starting to undo the very sharp loss in profits caused by that disaster (rebuilding industrial capacity isn't cheap, nor is it fast).

Moore never anticipated pre-flood manufacturing consolidation into one geographical region, either. Whether it's physical platters or flash memory, much of the actual manufacturing of hard drives is still stuck in a few small areas. That means rebuilding there is taxing construction capacity, it means that the specialized machinery needed to do everything from plating platters to testing flash memory is backlogged, and it means that there's still a somewhat credible manufacturing backlog.

Pricing and availability ARE better, by far, than in late 2011, but I agree with reader50 that it's not where it should be, though my idea of where prices should be is not quite so as his.

[Waragainstsleep]

Yes you're quite right, it was a flood. I think it was the RAM warehouses that caught fire. Or maybe a camera factory.

[akent35]

I actually was somewhat surprised at HD prices about 5 months ago, when I was looking to replace a defective 500 gig Seagate drive. I kept thinking that with most folks needing larger capacity drives, that prices for a 500 gig model would be cheap. But, was I ever surprised! I eventually purchased a 1 TB Seagate drive from Best Buy for $55, as 500 gig prices were almost the same. Recently, I did see an ad (on dealmac.com) for a 500 gig Seagate drive for $50!

I also suspect that for market share of HDs (non-SSD), it is as follows:

1. Seagate
2. Western Digital
3. Toshiba

In any event, at least SSD prices are coming down, though not fast enough.

[Mike Wuerthele]

We're not so speedy with deal prices all the time- I do try and get the ones I see on Fridays when I put the "weekly deals" posts together.

500 for $50 isn't super-hard. Microcenter has a Toshiba 500GB 3.5-inch drive for $46, with similar pricing for a 2.5-incher.

Toshiba 500GB 7,200 RPM SATA 6.0Gb/s 3.5" Internal Desktop Hard Drive - Bare Drive 965012 - Micro Center

[reader50]

Originally Posted by ghporter 

... but there is still the rather heavy load on consumer prices caused by the floods 3 years ago; vendors, distributors, and yes manufacturers, liked the higher prices, and they are really only just now starting to undo the very sharp loss in profits caused by that disaster (rebuilding industrial capacity isn't cheap, nor is it fast).

I've corrected some details in the quote. Right after the Thailand floods, the manufacturers predicted normal supply in 6 months. They got their fab capacity back online in 5 months, but found the higher prices were indeed very good. So they did a slow ramp-up, and took 18 months to return to pre-flood prices. Result? Record profits.

Western Digital stock moved from $25.26 to $53.29 (+3 dividends)
Seagate moved from $15.16 to $36.40 (+6 increasing dividends)

The duopoly has been sitting pretty ever since. All factories rebuilt, and money pouring out of the bank. It will continue until Toshiba ramps up their consumer offerings. Or until falling SSD prices become competitive for mass storage.

[ghporter]

I hadn't seen that they were back online so quickly. I still thought it took a much longer time to get all the equipment back into operation, and that there was a long backlog in getting new production equipment. I guess I believe too many advertising statements.

[akent35]

Originally Posted by EstaNightshift 

We're not so speedy with deal prices all the time- I do try and get the ones I see on Fridays when I put the "weekly deals" posts together.

500 for $50 isn't super-hard. Microcenter has a Toshiba 500GB 3.5-inch drive for $46, with similar pricing for a 2.5-incher.

Toshiba 500GB 7,200 RPM SATA 6.0Gb/s 3.5" Internal Desktop Hard Drive - Bare Drive 965012 - Micro Center

But, a 1 TB 7,200 RPM SATA 6.0Gb/s 3.5" Internal Desktop Drive (Bare drive) can be purchased for less than $65, even when they are not on sale. And, when they are on sale, one can get them for around $50 to $55.

Also, one needs to check almost daily for deals on hard drives. A good source is dealmac.com. That is where I saw the 1 TB drive on sale at Best Buy, and it was not a Friday (nor the weekend).

[Mike Wuerthele]

Originally Posted by akent35 

But, a 1 TB 7,200 RPM SATA 6.0Gb/s 3.5" Internal Desktop Drive (Bare drive) can be purchased for less than $65, even when they are not on sale. And, when they are on sale, one can get them for around $50 to $55.

True, but not my point.

[Spheric Harlot]

Originally Posted by ghporter 

I hadn't seen that they were back online so quickly. I still thought it took a much longer time to get all the equipment back into operation, and that there was a long backlog in getting new production equipment. I guess I believe too many advertising statements.

I was in sales at the time; it took almost a year until we had the same levels of stock as before the floods.

In the meantime, a lot of business started shifting to SSDs.

[akent35]

Originally Posted by EstaNightshift 

True, but not my point.

Understand. But, my point was that one can get a 1 TB drive (with just about the same specs) for about the same 500 gig sales price.

Also, one sees sales on 1 TB drives way more frequently than on 500 gig drives.

[Mike Wuerthele]

Originally Posted by akent35 

Understand. But, my point was that one can get a 1 TB drive (with just about the same specs) for about the same 500 gig sales price.

Also, one sees sales on 1 TB drives way more frequently than on 500 gig drives.

I knew what your point was

[P]

Originally Posted by akent35 

Understand. But, my point was that one can get a 1 TB drive (with just about the same specs) for about the same 500 gig sales price.

Also, one sees sales on 1 TB drives way more frequently than on 500 gig drives.

Hard drives are made up one or more platters. The capacity of each platter grows over time, but the number of platter does not increase especially (and has arguably gone down now that we're using thinner drives even on 3.5"). The capacity of one 3.5" platter passed 1TB while back, so that is essentially the cheapest drive you can make with current tech. Anything smaller is probably lower tech, which means that the cost to make them is essentially the same (possible exception is that the process for the smaller platters might not have to carry an investment anymore, but that will be tiny).

[reader50]

Originally Posted by reader50 

I meant, how long does the non-volatile SSD memory last without power? Unplug an SSD, set on shelf. How long can it sit before the data becomes uncertain? Your first quote suggests at least a year, and wikipedia also mentions in passing that SSDs can't hold data indefinitely without refresh power. But I haven't seen any hard data.

Originally Posted by P 

I know what you mean, but I don't know the answer. We have two data points:

* The JEDEC spec that a drive with exhausted write cycles should be readable for one year after that
* The ten year warranty that some enterprise drives have.

I just added a data point last night on flash memory storage persistence. I plugged in an old (~2001) answering machine, with solid-state memory. The outgoing recording was still there. The machine had been unplugged for 3.5 - 4 years.

The message sounded normal. Not as definitive as a hashcheck on a file, but it's something.