What do you all think of this...

[capeconsultant]

http://thessdreview.com/our-reviews...w-performance-meets-emlc-endurance-and-value/

??

Looks cool to me!

 

[Hulk]

Interesting find and a good read. Personally the 10k endurance for $1/Gb doesn't interest me. I have a feeling my 3k endurance drives will be working in 5 years based on what Anand has written and my personal experience with SSD's thus far. In addition , at the rate SSD performance is increasing and prices are decreasing I doubt I'll be using these SSD's for my main rig in even 2 years. I'll upgrade to something faster and use my current SSD's in secondary machines, for back ups, scratch disks,...

Now if you are someone who needs this type of endurance then this is a great buy I would think.

 

[capeconsultant]

And it comes right on the heels of this eMLC beauty http://thessdreview.com/our-reviews...w-performance-meets-emlc-endurance-and-value/

The SSD game is heating up DRAMATICALLY! Rock on!

 

[Revolution 11]

I am surprised no one offered eMLC drives to the consumer market before (if they did, oops). 2000/3000 cycles may be enough for 10 years but I have friends who won't touch SSDs because there are not enough write cycles.

 

[blastingcap]

There is more than one way to skin a cat, and there is more than one way to increase SSD endurance. I'm not sure which way is cheaper...

http://www.anandtech.com/show/6337/samsung-ssd-840-250gb-review/4

 

[Revolution 11]

I am interested in seeing how the Samsung tricks for keeping endurance high in TLC flash would increase the lifespan of MLC flash. That would be a nice increase in durability without too much price increases, TLC is not that much cheaper.

 

[capeconsultant]

TLC will never ever touch my computer. Really, we should all be using SLC, but for some reason, they say it is too expensive. Is it really???

 

[Zap]

TLC will never ever touch my computer. Really, we should all be using SLC, but for some reason, they say it is too expensive. Is it really???

Yes. Cost is related to how much silicon it takes, and SLC takes 3x the silicon of TLC for the same capacity.

 

[mrpiggy]

Problem with super NAND lifespan, is that the majority of SSD failures is due to controller/firmware failures, not running out of NAND write cycles. There are guys burning data through SSD's "attempting" to wear them out fast and it doesn't happen (at least to the good name brands). Current NAND lifespan is not an issue in consumer PC's with quality branded drives.

Check out these numbers on SSD's people are intentionally burning data through: http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm

 

[Hellhammer]

I am surprised no one offered eMLC drives to the consumer market before (if they did, oops). 2000/3000 cycles may be enough for 10 years but I have friends who won't touch SSDs because there are not enough write cycles.

Because eMLC has a downside: Shorter data retention time. Consumers don't usually write that much to their drives (current MLC is more than enough, even TLC is) but the drive may sit unused for long periods (an external drive, a longer vacation etc). If the timeframe is long enough, the SSD may have lost the data that was written to it.

Some of you might wonder why, but it's actually pretty simple. eMLC isn't much different from regular MLC, it simply uses smaller voltages for programming/erasing. That results in less stress for the silicon oxide but the charge is also weaker, which in turn it means it takes less time for the electrons to espace from the floating gate (i.e. shorter data retention time).

I also think eMLC has mostly been aimed for enterprises, which means NAND manufacturers will only use the highest quality NAND dies which are obviously more expensive.

Yes. Cost is related to how much silicon it takes, and SLC takes 3x the silicon of TLC for the same capacity.

Plus SLC enters production later than MLC, so while we have ~20nm MLC now, ~25nm SLC hasn't even been available for that long.

 

[Revolution 11]

How long do MLC SSDs hold data when unpowered? 5 years? 10 years? Does the time change when the rated cycles are reached (1000-3000 write cycles) vs when the drive has only a few cycles on it?

 

[Wall Street]

How long do MLC SSDs hold data when unpowered? 5 years? 10 years? Does the time change when the rated cycles are reached (1000-3000 write cycles) vs when the drive has only a few cycles on it?

That is one of those "theory vs. reality" questions that a lot of people have about SSDs. 20-nm class NAND didn't exist 5 or 10 years ago, so it is impossible to know though experimentation.

 

[Cerb]

How long do MLC SSDs hold data when unpowered? 5 years? 10 years? Does the time change when the rated cycles are reached (1000-3000 write cycles) vs when the drive has only a few cycles on it?

Some SSDs are rated for offline retention, but most aren't. Minimum retention could certainly be extrapolated for some given flash, with a few weeks or months of data across a wide enough range of chips, but most need to get their drives out to the market now, so the actual value is whatever chance has allowed for.

HDDs are basically the same way. I've personally seen CRC-verified backups "rot" on a shelf within several months. I think most flash is better than that, but we can't really be certain, right now, without manufacturers actually testing the worst-retention chips, and publishing those results as minimum specs.

 

[Revolution 11]

So if the drive is powered but no writes happen, does the data not disappear over time? Is the requirement here to keep the drive on or to write actively to it?

 

[Cerb]

So if the drive is powered but no writes happen, does the data not disappear over time? Is the requirement here to keep the drive on or to write actively to it?

With an HDD (except for a few, like WD AV and Red), you actually want to read the data.

With flash, I really don't know enough to know what happens if you have it on, but aren't writing to it. If you are writing, then it will be re-written as part of the wear-leveling process. Every read technically weakens it, and it loses charge over time, so each time it needs to be read, the drive might decide to move it as part of the next batch of writes.

If used regularly, it shouldn't be an issue, but the SSD makers are exceptionally guarded about their wear leveling specifics. I've tried to dig up details a few times, but the most I could find are some uni studies of single chips, or software-wear-leveled systems from several years ago.

 

[capeconsultant]

I agree about controllers causing the bulk of the issues. Is there anything we can do about that? Choose wisely???

 

[Hulk]

SSD's are a relatively new technology. Hard drives have had 50 years to mature. I'd bet most manufacturers have things pretty worked out now and things will only get better.

 

[blastingcap]

With an HDD (except for a few, like WD AV and Red), you actually want to read the data.

With flash, I really don't know enough to know what happens if you have it on, but aren't writing to it. If you are writing, then it will be re-written as part of the wear-leveling process. Every read technically weakens it, and it loses charge over time, so each time it needs to be read, the drive might decide to move it as part of the next batch of writes.

If used regularly, it shouldn't be an issue, but the SSD makers are exceptionally guarded about their wear leveling specifics. I've tried to dig up details a few times, but the most I could find are some uni studies of single chips, or software-wear-leveled systems from several years ago.

I imagine it can't be too hard to code some sort of conditional auto-scrub routine, but I could be wrong.

 

[Cerb]

I imagine it can't be too hard to code some sort of conditional auto-scrub routine, but I could be wrong.

It's not a matter of being hard. It's that the SSD manufacturers seem to consider such details to be among the secrets that they don't want the competition to be able to easy know, in case some aspect turns out to offer a competitive advantage or disadvantage.

Wear-leveling of data being read everybody does. Wear-leveling of data being written everybody does (obviously). Checking of recently-untouched data health, if the drive is severely under-utilized: ???

 

[Hellhammer]

How long do MLC SSDs hold data when unpowered? 5 years? 10 years? Does the time change when the rated cycles are reached (1000-3000 write cycles) vs when the drive has only a few cycles on it?

Around 1-3 months, although that rating is when the NAND has reached its P/E cycle rating. Fresh NAND has stronger silicon oxide, which leads to less leakage, but I'm not sure how long the data can last there with a new SSD.

So if the drive is powered but no writes happen, does the data not disappear over time? Is the requirement here to keep the drive on or to write actively to it?

When the drive is powered, the controller can do wear-leveling to keep the data there. Remember, even if you're mot writing to the drive, the controller may be doing writes in the background.

 

[Cerb]

I think some perspective may be good to add, here, too. This is now about very uncommon worst-case scenarios.

A drive with data on it being used will make any retention issues completely moot. The data lifetime is long enough that even a small amount of writes over time will be plenty for an on-demand-only static wear leveling system to take care of it all. If the drive is mounted, chances are well into the nines that it will be active enough that data won't be lost due to sitting untouched, even if a given drive does not actively perform background scrubbing of less-used blocks.

Then, the retention is related to the write cycles and temperature. IIRC, the chips are supposed to be rated for 1yr at the cycle count, but nothing published I can find easily is willing to go anywhere near that, even for normal consumer MLC--is that just CYA by drive and storage companies, or are there other issues? OTOH, retention on fairly new flash will be much higher than whatever the minimum is. You might get 10 years until you hit 1000 p/e, FI. Not only that, but recent controllers with their DSP magic (voltage over time analysis and control) can make the actual retain remain high for quite a long time, by reducing the actual wear.

The main issue we have today is that too much of it is unknown, without seeking out specific devices with specific ratings. SSDs as we have them now are still fairly new, and so are standards for offering guarantees for them as whole units. IIRC, the recent JEDEC standard requires 1yr (less for enterprise), so even if manufacturers start advertising compliance in their docs (I'm sure it will be just another footnote for all of them in a couple of years), you'll still have to seek out specific/esoteric products for higher offline guarantees.

So, don't fear the SSD, just don't trust it to sit in the back of a cabinet for 10 years and not lose any data, unless you know it is rated for that by the manufacturer.

 

[Revolution 11]

Since the information for SSDs is spotty, how about HDDs? I have two backup drives, each with the same data, that I backup onto irregularly from my desktop and two laptops. These drives sit in a dark, cool, dry location for several months at a time, unpowered. How long can I reasonably expect the drives to hold the data I need backups of?

I never actually read all the files on these drives, I just connect them to the necessary PC and copy files over. Once disconnected, back into storage. Would I just leave these drives plugged in for a few hours or does each file need to be read? The drives are used for a few hours at a time every few months so they are not lying completely inactive constantly.

It sounds more and more like archival DVDs are a better choice for medium-term storage (1-10 years).