HDDs prices

[nikalai2]

HDDs Shipments of PC Hard Drives Predicted to Drop By Nearly 50% in 2019

I know that this is not Malaysia flood but maybe it's possile to have higher pricess for spinner in the future!?

 

[cobrakiller58]

That makes sense to me, my main desktop does not have an HDD in it and most of my friends have no intention of getting HDD's in their new builds. I don't THINK datacenter and enterprise drives will increase in price as the demand for them is constantly increasing and multi TB SSD's are too expensive at the moment for bulk data storage.

 

[rvassar]

All my spinning rust disks are in my NAS, or sitting on a shelf.

SSD's are expensive, but... Consider we've had 4Tb SSD's available since at least 2014. Those were built from 32nm or 22nm process chips and cost upwards of $15k. The SATA SSD's got to 4Tb recently using 14nm chips, and cost what? Maybe $4k? We're currently moving 7nm wafers into production, and have 5nm & 3nm processes on the short roadmap (2024). Conversely 18Tb spinning rust disks are also imminent...

I'm thinking format change. The 2.5" disks for the mobile market will go away. They'll all be SSD's. The 3-1/2" segment is a wild card... I could see a change of format happening.

 

[cobrakiller58]

I'm thinking format change. The 2.5" disks for the mobile market will go away. They'll all be SSD's. The 3-1/2" segment is a wild card... I could see a change of format happening.

I could not agree more, I believe you can already purchase 30TB ssd in the 3.5" format if you have the appropriate connections and the funds.

 

[ngmaggie]

I could not agree more, I believe you can already purchase 30TB ssd in the 3.5" format if you have the appropriate connections and the funds.

30 TB SSD, is that available?

 

[cobrakiller58]

First I made a mistake it is a 2.5" drive Samsung PM1643 30.72 TB, availability I do not know personally probably special order only. I've never gone hunting for one because the price references I've seen were in the $12,000 range which is too rich for my blood but large corporations should be able to get their hands on them.
Edit:
lol just put the part number (MZILT30THMLA) in tiger direct and you can place an order none in stock obviously
http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=5971675

 

[rvassar]

First I made a mistake it is a 2.5" drive Samsung PM1643 30.72 TB, availability I do not know personally probably special order only. I've never gone hunting for one because the price references I've seen were in the $12,000 range which is too rich for my blood but large corporations should be able to get their hands on them.
Edit:
lol just put the part number (MZILT30THMLA) in tiger direct and you can place an order none in stock obviously
http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=5971675

Wow! And look at that $19k price tag...

 

[Nassiary]

It was $15 last year,i got one.

 

[jgreco]

HDDs Shipments of PC Hard Drives Predicted to Drop By Nearly 50% in 2019

I know that this is not Malaysia flood but maybe it's possile to have higher pricess for spinner in the future!?

Yes, the general theme here is that the days of the conventional HDD are numbered. With shipments falling precipitously, drive manufacturers have to make their money somewhere, so they are focusing on capacity, and probably going to have to hold the line on price per TB.

Flash is coming to eat their lunch. 16TB/18TB HDD's are the largest available, and density improvements, though frequently promised by manufacturers, have been slow in arriving. Seagate's been talking about 40TB+ for awhile.

Meanwhile, flash prices are once again rapidly falling, and it isn't difficult to fit a bunch of flash into a small package.

In 2011, a consumer grade 240GB SSD cost $440, or to normalize to cost/TB, around $1760/TB.

In 2019, an inexpensive 1TB SSD costs around $110.

So in 8 years, prices have dropped to about 1/16th.

Right now, the cheapest HDD is generally something like the shucked 8TB HDD's available from time to time for $129, or $16/TB. Back in 2012, you could shuck 3TB HDD's for a bit less than that, $109, or $36/TB.

So in 7 years, prices have dropped a bit more than half.

Right now, if I wanted to replace 8TB shucked HDD's with 8TB SSD's, hypothetically, based on 4TB SSD prices, I would expect maybe $1200 for that SSD device. That is 10x the cost of the 8TB shucked HDD. This should make it clear that this is a nonstarter today.

But over the next five to ten years? I think we'll start to see flash continue to eat into the HDD market. HDD isn't evolving as fast. SSD will catch up.

 

[cobrakiller58]

I was looking at the cost of 15K HDD's per TB and they make the PM1643 not look so expensive however if you need 15K drives you will likely just get some cheaper faster NVME SSD's instead.
Are there any case's where 15k HDD is superior choice?

 

[jgreco]

I was looking at the cost of 15K HDD's per TB and they make the PM1643 not look so expensive however if you need 15K drives you will likely just get some cheaper faster NVME SSD's instead.
Are there any case's where 15k HDD is superior choice?

Yes, anywhere where you are beating the living crap out of the drives, such as a Usenet transit box, where traffic levels are measured in drive writes per day even for very large drives. The interesting thing is that for high volume applications where you end up using smaller drives, you can end up with dozens or even hundreds of drive writes per day.

 

[rvassar]

Yes, anywhere where you are beating the living crap out of the drives, such as a Usenet transit box, where traffic levels are measured in drive writes per day even for very large drives. The interesting thing is that for high volume applications where you end up using smaller drives, you can end up with dozens or even hundreds of drive writes per day.

Actually... I'd argue that kind of use case has already gone to Flash where the funding allows. That was kind of the basis for the HGST (Virident) FlashMax II cards. These were 550Gb to 4.8Tb PCIe flash memory cards that were in production 2014 - 2015. They had massive flash oversubscription, almost 50%. They were rated for continuous write at 2.8Gb/sec for 5 straight years, which works out to 50 drive writes per day for the 4.8Tb model. The primary consumers were large Internet properties that needed fast content access & caching, very much like a Usenet transit box, but with money to spend.

 

[jgreco]

Actually... I'd argue that kind of use case has already gone to Flash where the funding allows. That was kind of the basis for the HGST (Virident) FlashMax II cards. These were 550Gb to 4.8Tb PCIe flash memory cards that were in production 2014 - 2015. They had massive flash oversubscription, almost 50%. They were rated for continuous write at 2.8Gb/sec for 5 straight years, which works out to 50 drive writes per day for the 4.8Tb model. The primary consumers were large Internet properties that needed fast content access & caching, very much like a Usenet transit box, but with money to spend.

I'm not aware of anyone doing this. I both develop a major Usenet news software package and run multiple transit servers. Speaking as someone who's been doing this for several decades, most Usenet companies are interested in cheap. A pile of cheap HDD has more endurance - probably even on a per-spindle basis. The last transit I deployed, we literally went through the shop scrounging up whatever random HDD's we could find that didn't seem to have a better purpose in life, and we put them in and are letting them get pounded to death.

 

[rvassar]

I'm not aware of anyone doing this. I both develop a major Usenet news software package and run multiple transit servers. Speaking as someone who's been doing this for several decades, most Usenet companies are interested in cheap. A pile of cheap HDD has more endurance - probably even on a per-spindle basis. The last transit I deployed, we literally went through the shop scrounging up whatever random HDD's we could find that didn't seem to have a better purpose in life, and we put them in and are letting them get pounded to death.

You missed my point "where funding allows"... Commercial applications that require high drive write counts have already gone SSD. These are like Usenet transit servers in I/O abuse rate, but obviously not Usenet transit sites. Think large Internet properties, social media sites, large high traffic shopping, etc... They're the one's that were paying for those $14k PCIe SSD cards.

 

[jgreco]

You missed my point "where funding allows"... Commercial applications that require high drive write counts have already gone SSD.

No, the commercial applications that require HIGH drive write counts per day have not, especially if they're just trying to find affordable tech to solve the problem. A HDD can do ingest at 100MBytes/sec which is 9TB/day (higher if you're doing minimal to no seeking) which works out to 16PBW over 5 years. Your high end Optane 900P is rated at 8.8PBW. The high end FlashMax II 1.1TB was rated at 16PBW. But at what cost? And the thing is, we put *24* HDD's into a transit server which is how we fill 40Gbps pipes with these things. There's no way to get 24 FlashMax II's into a server...

These are like Usenet transit servers in I/O abuse rate, but obviously not Usenet transit sites. Think large Internet properties, social media sites, large high traffic shopping, etc... They're the one's that were paying for those $14k PCIe SSD cards.

This is kinda pointless... there have long been places with budget that have paid to do stuff like this with the fastest tech, which usually is "in RAM", which also happens to have unlimited endurance. In the context of the question that was being asked, though, we were talking about the *low* end of the budgetary realm, and where 15K drives might still be (at least for a little while longer) dominant. I think I gave a reasonable use case where that still works out.

The "do it in RAM" answer is at the stupid-high-end of the cost spectrum and the "do it on HDD" is still on the low. Both have basically unlimited endurance.

I certainly buy that the day is coming when SSD kills the HDD market.

 

[rvassar]

This is kinda pointless...

I certainly buy that the day is coming when SSD kills the HDD market.

Agreed... But I never seem to get job offers from the "do it in RAM" types... ;)

 

[jgreco]

Agreed... But I never seem to get job offers from the "do it in RAM" types... ;)

Well if it makes you feel better, it wasn't so long ago (only a bit more than 20 years) where one of my customers needed to order a P/E-P55T2P4D and loaded it up with eight custom-manufactured 64MB SIMM's (512MB RAM), which cost a ridiculous amount of money because the chips on them had only been on the market for a month or so.

So if you look at it like that, most of the jobs we used to "do in RAM" at a shocking price premium years ago are now done in RAM without even thinking about it. It's all about the scale, eh.

 

[cobrakiller58]

Interestingly, I tried to find Seagate's workload limit on their 15K drives and it didn't seem to be listed anywhere. The enterprise ones I'm using are rated at 1.5TB/day.... their warranties would be void in no time using them in the manner you two are describing.

 

[jgreco]

Interestingly, I tried to find Seagate's workload limit on their 15K drives and it didn't seem to be listed anywhere. The enterprise ones I'm using are rated at 1.5TB/day.... their warranties would be void in no time using them in the manner you two are describing.

"Workload limits" are a brand new thing that only appeared within the last few years as HDD marketing departments tried to justify the massive cost differences between relatively similar mechanisms (taking a page from the SSD marketing world). Well, to be charitable, the introduction of SMR had something to do with it too. Rewind even five years and ask about "hard drive workload limits" to a storage professional and they'd have looked at you like you were on the worst sort of drugs. Back in the day no one would ever have been concerned about a 100% write duty cycle.

 

[cobrakiller58]

Having it in my mind is a good way to reject warranty claims, all they have to do is pull the smart data from the drive and compare the power on hours and how much has been written to it.