SSD's got even bigger? PLC nand

[Mihle]

As long as a SSD is 300 MB/s+ and there is big enough cache for writes, its fast enough for me.

Slower than that it would only really be worth it if the price gets very similar to HDDs, OR they are less likely to fail than HDDs.

[captain_to_fire]

1 hour ago, Origami Cactus said:

QLC is already really slow at only 80-120mb/s for sustained writes, so this will be even slower.

This reminds me on why QLC is slow 

 

[Zodiark1593]

1 hour ago, VegetableStu said:

Hitmonlee, Hitmonchan, and Hitmontop

Bruce Lee, Jackie Chan, ???

[Tristerin]

When recovering offal matters

[NDRE28]

12 hours ago, Origami Cactus said:

https://arstechnica.com/gadgets/2019/09/new-intel-toshiba-ssd-technologies-squeeze-more-bits-into-each-cell/?amp=1

 

 

 

My thoughts:

 

Hope it doesn't make it's way into consumer products any time soon. QLC is already really slow at only 80-120mb/s for sustained writes, so this will be even slower.

Also it doesn't benefit that much:

Space increase per cell increase:

SLC 

MLC +100%

TLC +50%

QLC +25%

PLC +12.5%?

 

Also another question is durability.

 

I would only consider PLC, if you could get it in 6+TB capacities, and for the same price as hdd.

Because it will be slower than a hdd, but atleast it will be fully silent.

I don’t go lower than TLC! 

I have 2 MLC drives and a TLC (NVMe) drive. 

The next drive that I’ll buy will be NVMe MLC. 

 

PS:

Big, fast and cheap cannot exist! Novices will buy PLC NVMe drives thinking that it’s a good deal!

[Zodiark1593]

5 minutes ago, VegetableStu said:

i believe that was how those two pokemon were named? o_o

Would hardly know it just by looking at the Pokemon.

 

While were on the topic, I'm quite curious as to who on the English translation team thought that Snake spelled backwards made a fine name.

[Kisai]

2 hours ago, 5x5 said:

This is literally the point where an HDD makes more sense as a product. Who in their right mind thinks making an SSD slower than a 5400RPM HDD is a good idea?

This has at least one use case:

 

Tape drives.

 

Have you see how big a tape drive array is today? I'll give you a hint. A LTO-8 tape is 12TB and $260 each. A Quantum Scaler library is about $10,000 without the drives, and about $4000 per drive, so to have a fully functional system you're looking at probably at least two drives 50 tapes for 600TB of storage. So the entry level price is around $31000, or a little over $50/TB, compare that to consumer SSD's:

Samsung QVO $140 1TB

Samsung EVO $240 1TB

Samsung PRO $300 1TB

 

The office has a guy come in once a week to pull tapes to put into long term storage and replace them with blanks. So let's assume for a minute that he replaces 10 tapes a week, they probably spend $130,000 on tapes per year, at this office alone. 

 

Now, let's assume that 5LC costs... oh $50/TB, That puts it at a price point where it's competitive with tape. And unlike tape drives, they don't take up a huge amount of space, have no moving parts, and don't really need to be written to more than once to be useful. So let's say we need 12 1TB SSD's to equal one tape, hmm where can we fit 500 SSD's.... SuperMicro has the X11 platform that can take 32 drives of EDSFF format SSD's. 

 

So Intel sells their 8TB TLC SDSFF's for $1800/ea (225$/TB), which is right on par with Samsung EVO (TLC)'s cost per TV. So if 32 of these can be loaded in a server, 256TB are available in one machine. We're still 44TB short of the tape drive configuration, so for the sake of price let's make that a full 600TB with 75 drives. 75 x 1800 = $135,000, now let's use intel's 660p data instead, $129/TB, $77,400. Presumably we could get to $50/TB with 5LC, making it $30,000.  Now how much does the server cost? CSE-136HTS-R1K69P-E2 costs $4500. 

 

So initial investment in theory is about $34,500 for 75 5LC SSD's+one server, Initial investment with the Quantum Scaler is about $31000 (including first 50 tapes.) The SuperMicro server is a 2U, where as the tape drive/library is 3U for the configuration described above. The initial investment is much higher for the tape system, regardless of tapes and drive configuration used. The ongoing cost of 50 x 12 vs 75 x 8 might be more interesting to look at. So as I said above, I assume the guy comes in once per week to swap out 10 tapes. So if the cost per TB is the same, then 120TB is being cycled out per week, which is a lot slower with the tape units since the robotic arm has to fetch the tapes one (or two) at a time. 15 drives would be pulled per week instead and since they're nowhere near as fragile as the tapes (the tape drive library has been broken at least twice in the last year) they can just be ejected, placed in a ESD case, doesn't need to be magnetically shielded, and probably takes up about the same amount of physical space.

 

Of course this is all theoretical. If you have all these drives online at once, you can write to all the drives simultaneously for a backup, thus the network bandwidth would be saturated long before the drive performance becomes an issue. So the backup time starts to become a non-issue.

 

Anyway. Just for comparison sakes:

Spindle of 50 25GB BD-R's = $37USD/1.25TB ( $30/TB)

Spindle of 25 BDXL M-Disc 100GB = $250/2.5TB ( $100/TB)

 

Presumably M-Disc is supposed to be indestructible. But people aren't doing their enterprise backups to those.

 

[yolosnail]

Why are people making out like this is a bad thing?

 

You do realise that the average person does not care in the slightest how fast their hard drive is, they care about how much they can store on it. And as long as SSDs store less than a HDD for the same price, the average person will still buy a HDD instead.

 

Any enthusiast is likely always going to have a primary high speed drive for their OS install, and a secondary larger drive for their games and 'stuff', which these types of drives are perfect for. 

 

Heck, even some low end laptops ship with a small NVME drive for the OS and a 1-2tb HDD for mass storage, even with the 'lower reliability' of PLC NAND, it's going to be a heck of a lot more reliable than spinning rust in a laptop that gets knocked around all day!

[Donut417]

Is there a technical reason they cant just build SSD's bigger. Like 3.5 inch like standard hard disks size?

[leadeater]

17 minutes ago, Kisai said:

Have you see how big a tape drive array is today? I'll give you a hint. A LTO-8 tape is 12TB and $260 each. A Quantum Scaler library is about $10,000 without the drives, and about $4000 per drive, so to have a fully functional system you're looking at probably at least two drives 50 tapes for 600TB of storage. So the entry level price is around $31000

Actual prices are way lower than this, buy in price for LTO-7 when we did that for an HPE MSL6480 Base unit with 2 additional expansion modules and 240 LTO-7 tapes was around $30k NZD (19k USD).

 

Before that I was using two Quantum i80 libraries and just slightly over 2500 tapes, honestly needed better units than these for that many but we didn't start with that many.

 

17 minutes ago, Kisai said:

The office has a guy come in once a week to pull tapes to put into long term storage and replace them with blanks. So let's assume for a minute that he replaces 10 tapes a week, they probably spend $130,000 on tapes per year, at this office alone. 

Normally tapes go through a rotation cycle and get reused.

 

Raw cost wise nothing beats tapes for $/GB still.

 

Edit:

17 minutes ago, Kisai said:

which is a lot slower with the tape units since the robotic arm has to fetch the tapes one (or two) at a time.

You just pull the entire tape tray out, nobody uses the mail slots to remove a full rotation of tapes. Those are for small operations like take out 2 tapes or putting 2 in for recovery, even then I just release the whole tray anyway as the library can operate while it's open with the tapes already in the drive and you shouldn't take more than 2-5 minutes doing it.

[leadeater]

4 minutes ago, Donut417 said:

Is there a technical reason they cant just build SSD's bigger. Like 3.5 inch like standard hard disks size?

Like these? https://www.datacenterdynamics.com/news/nimbus-data-squeezes-100tb-into-a-35-inch-ssd/

[Doobeedoo]

Yeah, the so called lower quality type of SSD is not what I had in mind as 'SSD to get cheaper. Really I want to see MLC to come down in price, I'm mainly interested in those. 

[Lady Fitzgerald]

3 hours ago, Origami Cactus said:

...So the differences between different voltage states is much smaller, making the drive more error prone, and reducing the life span. Also if the drive is without power, it will loose it's data even faster than QLC.

Yes it has it's use cases, but i would like to see a cheap 8TB QLC ssd before that.

 

Those are my three largest concerns about QLC and PLC: more error prone, reduced life span, and poorer data retention when powered down.

 

While dramatically overkill, I make sure my backup drives get powered up and read at least once a month, especially the ones that have static data on them, to avoid the danger of corrupting or losing data due to charge leakage during downtime. It's not hard for me to do since i just run a backup every month, even if there is nothing new to backup, just before swapping the onsite backups with the offsite backups, then run the backups again on the formerly offsite backup drives when I get them home. It takes very little time on my part, especially since the computer does most of the work for me. I can update all ten 4TB backup drives while watching the news on TV, most of the time spent actually watching the news while the computer does its thing.

3 hours ago, The Torrent said:

Tlc is as low as I will go, and even then I’ll never fill over 50% capacity.

TLC is as low as I'll go for now. However, keep in mind that, while the early TLCs were a bit of a disaster and I wouldn't go lower than MLC, TLC has improved considerably and I have no problem using high quality TLC drives for my backup drives which, generally speaking, get far less usage than my data drives (which are all MLC; overkill but I like a large safety margin).

 

You are wise to not overfill your drives (most people simply cannot grasp the need for free space and overprovisioning, often confusing one with the other; they are not the same) but 50% is overkill for TLC and MLC. If the drive is getting written to and/or files are being deleted frequently, 25% free space (in addition to factory overprovisioning; factory overprovisioning should never be tampered with) should be plenty. If the drives are getting mostly read activity or are usually idle (few writes and deletions), 20% will probably be fine.

 

3 hours ago, 5x5 said:

This is literally the point where an HDD makes more sense as a product. Who in their right mind thinks making an SSD slower than a 5400RPM HDD is a good idea?

There are situations where having an SSD that is slower than an HDD can make sense. Increased speed is not the only advantage that SSDs have. Others include lower weight, smaller size, lower power consumption, less noise and heat generated, etc. (edit: forgot to mention resiliance to physical shock; thanks to @XenosTech for reminding me). This could be huge for laptops.

 

Even though it cost a small fortune (and means I'll be driving my 10 year old truck for probably another ten years), the biggest advantage for me switching to all SSDs was reduced weight and size. The weight of lugging multiple backup HDDs to and from my safe deposit box at my credit union every month was killing my shoulders (and my back wasn't all that thrilled, either) so the reduction in weight was a huge advantage. Size reduction was also good since I was running low on space in my safe deposit box and the space needed for them at home was becoming an issue.

 

The increased speed, lower heat generation, and reduced power draw were just bonuses for me (not that I didn't welcome them with open arms). TLC was the lowest available at the time and I'm sure as heck not going to trade them in for QLC. I won't even replace any of my MLC or TLC drives that might fail (I've had only one SSD fail on me so far; a 128GB Samsung 840 Pro that almost made it to the five year mark) with QLC until QLC has improved to the point they are at least as good as the current TLCs.

 

3 hours ago, strajk- said:

Well, it's good for archiving, so I'm happy on that end...

Actually, unless the drive is powered up several times a year or is kept powered up, any SSD is not good for archival since they willl eventually lose data while setting on the shelf; HDDs are far more suitable for cold storage (and even they have to be "exercised" every now and then). Of course, the more bits per cell, the sooner data can leak away.

1 hour ago, Donut417 said:

Is there a technical reason they cant just build SSD's bigger. Like 3.5 inch like standard hard disks size?

The article in the post below is an example that there is no technical reason they can't be built bigger. These will be a boon for server farms to increase capacity without building more infrastructure. Once they can trickle down to the consumer level (not likely but we can dream), they will allow us to have huge amounts of data in far less physical space than even current SSDs provide without the need for umpteen SATA ports or PCI-e slots.

1 hour ago, leadeater said:

Like these? https://www.datacenterdynamics.com/news/nimbus-data-squeezes-100tb-into-a-35-inch-ssd/

I'm glad to see someone finally brought one to market. Seagate teased us with one several years ago but never delivered. Imagine; 100TB in a space that currently yields less than 20TB.

[Origami Cactus]

10 minutes ago, Lady Fitzgerald said:

While dramatically overkill, I make sure my backup drives get powered up and read at least once a month,

Wouldn't be overkill with an QLC or PLC drive, more of a necessity.

[Lady Fitzgerald]

1 minute ago, Origami Cactus said:

Wouldn't be overkill with an QLC or PLC drive, more of a necessity.

You're probably right.

[XenosTech]

3 hours ago, leadeater said:

Even if the reads are only 200MB/s-250MB/s (which is lower than I expect) that's still faster than an HDD and the access latency times will be below ms to less than 3ms (bar cases where it makes no difference) so will still be much faster than an HDD, even a good 7200 rpm.

 

Remember back to the first generation of SSDs that were barely faster than WD VelociRaptors of the time? Usability and feel of those SSDs were still much better, no seek times = better.

Let's not forget we can cram more of these in a server than we can HDDs AND we can also chuck them in a shopping cart and take them across the parking lot to the new building without data loss lol

[Donut417]

35 minutes ago, Lady Fitzgerald said:

they will allow us to have huge amounts of data in far less physical space

At the cost of performance and reliability. 

[Lady Fitzgerald]

8 minutes ago, VegetableStu said:

... you know what, I'm actually starting to warm up on the idea of using 5LC SSDs for cold storage. maybe thaw them twice per year just in case

Being a confirmed coward, I would "thaw" (I like that term) more often than that.

[Lady Fitzgerald]

Just now, Donut417 said:

At the cost of performance and reliability. 

Why would a larger capacity SSD have reduced performance and reliability unless it was QLC or PLC? Being an enterprise drive, I'm thinking that drive will be SLC or MLC, possibly TLC, but almost definitely not QLC.

[Donut417]

Just now, Lady Fitzgerald said:

Why would a larger capacity SSD have reduced performance and reliability unless it was QLC or PLC? Being an enterprise drive, I'm thinking that drive will be SLC or MLC, possibly TLC, but almost definitely not QLC.

That’s what the tread is pertaining to PLC. At this point 3.5 inch SSD’s for larger storage. 

[The King of the Undead]

2 hours ago, Zodiark1593 said:

Bruce Lee, Jackie Chan, ???

stan lee?

[Lady Fitzgerald]

7 minutes ago, Donut417 said:

That’s what the tread is pertaining to PLC. At this point 3.5 inch SSD’s for larger storage. 

First of all, the term is "thread", not "tread". Second, Donut417's query didn't need to be limited to just PLC drives. threads often go off topic a bit (even a lot at times); build a bridge and get over it.

 

I was doing some research on the ExaDrive100 and it uses MLC NAND (researching this was not difficult; you should try it sometime). The manufacturer is claiming speed and durability to be equal to or better than current, smaller enterprise SSDs but its biggest claim to fame include reduced power consumption over even current enterprise SSDs and the ability to use existing infrasturcture without the need for adapters; just yank an HDD and pop this puppy in.

[Origami Cactus]

42 minutes ago, VegetableStu said:

... you know what, I'm actually starting to warm up on the idea of using 5LC SSDs for cold storage. maybe thaw them twice per year just in case

QLC is already not suitable for cold storage because of the difference between voltage levels being very small, PLC would need to be updated once a month atleast with current error correction technology.

[Zodiark1593]

25 minutes ago, will4623 said:

stan lee?

Hitmonstan?

 

Just now, Origami Cactus said:

QLC is already not suitable for cold storage because of the difference between voltage levels being very small, PLC would need to be updated once a month atleast with current error correction technology.

I wonder how much power is required to perform the refresh? Could have space for a battery to perform this task periodically.

[Origami Cactus]

7 minutes ago, VegetableStu said:

aaaaaaand into the trash that idea goes ,_,

 

(how about 3LC?)

 

at that rate you might as well stand an SSD on its shorter edge and call it a day ._.

3LC only has 8 levels of voltage per CELL, still much more than MLC and SLC, but it should last much longer than PLC (32 voltage levels), you have to keep in mind that TLC is pretty old at that point, so much more advanced than first gen QLC.