SSDs and Reliability

[jsmith65]

I've been reading a lot lately about SSDs and two things seem very obvious:

1. They're insanely fast.
2. No one seems to be super sure how well they'll perform in the long run.

I've read all over anandtech and other tech sites and just cannot find any solid information on how well drives will retain their performance over time when put in RAID 0 configuration where TRIM is not applicable.

Now in a little less than a month I'll have the money to buy the final components of my build. I've held off on the boot drives because of this lack of assurance. Right now I'm looking at either the 240GB Revodrive X2, two OCZ Vertex 2 120GBs in RAID 0, or 2 WD Velociraptor 600GBs in RAID 0.

Should I be worried about the SSDs degrading or failing? How does the reliability of modern SDDs compare to the WD Velociraptor? Are there other ways, besides TRIM, to maintain SSDs that are in a RAID configuration? Any help would be greatly appreciated.

 

[VirtualLarry]

If I were to get an SSD these days, it would be an Intel 320 drive. They have the 120GB version for $210 at Newegg after a coupon code.

 

[Voo]

From all we know about return rates and published failure rates we can say the following:

SSDs and HDDs have about the same return rates but:
- 7.2krpm 2tb+ HDDs are off the charts and extremely unreliable
- Intel SSDs are the shining example out there with significantly lower return/failure rates than anything else


Now if we're talking about performance degradation in RAID setups there are tests for that too - it basically comes down to how good the GC works.

 

[sequoia464]

I have had three sets of SSD's in raid0 without any perceptible issues for a year now - no problems with any of them.

2 Intel 40 GB 25V's
2 Intel 80 GB 25M's
2 OCZ 60 GB Vertex 2's

I am by no means very knowledgable, I have just gone off of others recomendations and have had no problems in raid0.

One of the things that has been stressed is to maintain some free space on the drive for the internal cleanup to be able to function efficiently. I am using 102 GB of the 120 available on the OCZ's and 130 of the 160 available for the Intel 25M's.

I also backup nightly just in case - that way if I do have a slowdown I can do the secure erase and then good as new.

 

[ncalipari]

2 Intel 40 GB 25V's
2 Intel 80 GB 25M's
2 OCZ 60 GB Vertex 2's

I am by no means very knowledgable, I have just gone off of others recomendations and have had no problems in raid0. .

Your raids works just fine, the problem is that raid further degrade the cells of the SSD preventing the automatic cleanup mechanism (TRIM) to work....

This means that they will die before non Raid-ed SSDs..

 

[nanaki333]

http://www.ocztechnologyforum.com/f...-A-simple-guide-for-speeding-up-EOL-OCZ-SSD-s.

tony trim helps clean up raided SSDs

 

[Drsignguy]

Other than a DOA drive. Or a possible firmware issue that can be updated, I think reliability all depends on the end user. If you use the drive like you normally should, then you should expect them to last a long time. You start abusing it, then the drive would most likely start degrading. I think we worry too much...

 

[sequoia464]

Your raids works just fine, the problem is that raid further degrade the cells of the SSD preventing the automatic cleanup mechanism (TRIM) to work....

This means that they will die before non Raid-ed SSDs..

Absolutely true that trim does not work in raid. Every modern SSD controller however has their own 'garbage collection'. This is what works to 'trim' the drives if they are in a non Windows 7 environment. This is why Mac's can still use SSD's - especially some such as the Kingstons that have aggressive 'garbage collection' allogorithms.

Your comment is the first that I have ever seen about raided SSD's dieing faster than non raided SSD's. I'm not certain that it it is accurate.

Perhaps others can comment on this

 

[RU482]

reliability in SSDs.

Follow these rules.

1. Avoid re-branders. If they don't have a hand in making the NAND, then say NAH
2. see rule 1

PS...Indillinx sucks

RE: I have purchased approx 500 Intel (G1 and G2), 200 Samsung, 100 Crucial/Micron, 300 Kingston, 500 OCZ, and 800 Supertalent Drives for my company. Guess which 1600 have given me the most trouble.

 

[RU482]

Other than a DOA drive. Or a possible firmware issue that can be updated, I think reliability all depends on the end user. If you use the drive like you normally should, then you should expect them to last a long time. You start abusing it, then the drive would most likely start degrading. I think we worry too much...

what in the bloody hell are you talking about "abusing" a drive? What does that even mean? Wear ratings on NAND are so far out of the reach of even the extreme home user... your comment is ridiculous.

 

[Elixer]

reliability in SSDs.

Follow these rules.

1. Avoid re-branders. If they don't have a hand in making the NAND, then say NAH
2. see rule 1

PS...Indillinx sucks

RE: I have purchased approx 500 Intel (G1 and G2), 200 Samsung, 100 Crucial/Micron, 300 Kingston, 500 OCZ, and 800 Supertalent Drives for my company. Guess which 1600 have given me the most trouble.

Quite a few SSDs you got there... you got any failure rate info for specific models ?
Do most of the failures just die, or do they throw SMART errors before they die?

 

[Voo]

Follow these rules.

1. Avoid re-branders. If they don't have a hand in making the NAND, then say NAH
2. see rule 1

Soo.. if Intel is using IMFT flash in their products then its perfectly fine, but if OCZ uses exactly the same flash then suddenly the flash is horrible? Intriguing logic - say how many of your drive failures where due to flash? Because the source for the vast majority of problems is fw/controller..

 

[RU482]

Soo.. if Intel is using IMFT flash in their products then its perfectly fine, but if OCZ uses exactly the same flash then suddenly the flash is horrible? Intriguing logic - say how many of your drive failures where due to flash? Because the source for the vast majority of problems is fw/controller..

Not quite what I am saying. A company like OCZ doesn't put out as consistent of a product as Intel. You'll get a batch with zero problems, then you'll get a batch with abnormal failure rates.

If a company has a hand in the manufacture of the components, they likely have more and better resources that result in better quality and better quality control. They also maintain a more consistent BOM than a company that is buying flash on the open market.

my $0.02

 

[wanderer27]

From an Article on the Register:

As for reliability concerns, these are answered in the old-fashioned way: make lots of backup copies. Kingston says a typical MLC unit has a write/erase cycle endurance of approximately 10,000 per cell, compared to 100,000 cycles per SLC cell, which means SSDs use a number of different techniques to spread the wear across the drive.

I'd love to get one myself, but price and lifetime are my biggest concern right now.


.

 

[corkyg]

This thread speaks a lot to SSD reliability:

http://forums.anandtech.com/showthread.php?t=2160472

Especially the groberts101 thread near the end - such lockups do weigh on reliability.

 

[bradley]

Companies like OCZ try to make the cheapest products possible, including whatever NAND is available at hand. The many NAND varieties make their products harder to support and give users a huge learning curve. Though destructive flashes usually solve most problems.

Even with Intel producing everything in-house with a singular controller and NAND, they've had their share of problems. Still, I would stick with Intel if you want reliability right out of the box.

 

[Voo]

Not quite what I am saying. A company like OCZ doesn't put out as consistent of a product as Intel. You'll get a batch with zero problems, then you'll get a batch with abnormal failure rates.

If a company has a hand in the manufacture of the components, they likely have more and better resources that result in better quality and better quality control. They also maintain a more consistent BOM than a company that is buying flash on the open market.

my $0.02

That would still imply that the majority of failing SSDs is because of flash problems. Which I totally don't see, actually I can't remember anyone here on the whole forum that had problems with failing flash..

@bradley: I'd love to see how NAND varieties make for a larger user learning curve (user learning curve and SSDs??). And can you remember any firmware problem with SF drives that only applied to only one kind of nand flash? I don't but I assume it could be possible. Seems more like rare flash leads to fewer optimizations and therefore lower speed (one thing that was mentioned in one of anand's last reviews) but that seems about it. Although I assume it's possible that the incompatibilities are partially because of problems with all the different flash and a leaky abstraction layer - impossible to say with what we know I think.

 

[smakme7757]

I was unlucky and had my first SSD die after a month. It just wouldn't be dected by the BIOS and that was that. But with that being said a duplicate of any important file or files is important anyway - Take a backup!

Both SSDs and mechanical HDDs can fail, at any time, so a backup is extremely important! But i (personally) feel mehanical HDD are more reliable and easier to fix when something goes wrong.

 

[bradley]

That would still imply that the majority of failing SSDs is because of flash problems. Which I totally don't see, actually I can't remember anyone here on the whole forum that had problems with failing flash..

@bradley: I'd love to see how NAND varieties make for a larger user learning curve (user learning curve and SSDs??). And can you remember any firmware problem with SF drives that only applied to only one kind of nand flash? I don't but I assume it could be possible. Seems more like rare flash leads to fewer optimizations and therefore lower speed (one thing that was mentioned in one of anand's last reviews) but that seems about it. Although I assume it's possible that the incompatibilities are partially because of problems with all the different flash and a leaky abstraction layer - impossible to say with what we know I think.

The controller adjusts for all sorts of variables, including latencies. OCZ purchases NAND from different manufacturers, and likely batches that were binned differently. We are talking about the bottom of the barrel here. Flash procedures change based on NAND type. For constancy, buy in-house brands like Intel and McDonalds, not OCZ and the local diner. For the record, my Solid 2 (with Intel NAND) is going strong after almost 1.5 yrs.

 

[Voo]

The controller adjusts for all sorts of variables, including latencies. OCZ purchases NAND from different manufacturers, and likely batches that were binned differently. We are talking about the bottom of the barrel here. Flash procedures change based on NAND type. For constancy, buy in-house brands like Intel and McDonalds, not OCZ and the local diner. For the record, my Solid 2 (with Intel NAND) is going strong after almost 1.5 yrs.

Well and do you know of any case where a SF drive failed because of the flash's PE cycles? I think those problems are a small minority and most problems seem to be related to compatibility between the FW and MB - eg drives just locking up in Macs.

Most of the flash in SF drives is IMFT anyways, i.e. coming from exactly the same fab as the flash in Intel drives - even with the same specs. So maybe the IMFT flash is good for 10% more PE cycles (and there's no prove whatsoever for that), but as long as its good for the specified 3k PE cycles that's hardly a problem. The only problem I see is using flash with noticeable different performance without chaning the model number and OCZ does something about it as they should.
So as long as nobody can show a larger sample of drives with faulty flash that failed before its 3k PE cycles, that's just FUD in my book~

 

[bradley]

Instead of the marketing literature, search on forum anecdotes, not only here but @ Intel and OCZ. I've seen enough examples of SSD failures attributable to just about every cause.

In fact, I believe there are many more misrepresentations by NAND & SSD manufacturers than just about any other industry. Look how many people still believe a shrink to 25nm means faster speeds and greater longevity, when it really only benefits the manufacturer's bottom line.

 

[bradley]

The only problem I see is using flash with noticeable different performance without chaning the model number and OCZ does something about it as they should.

On this, I agree with you 100 percent. It's the reason I will no longer purchase OCZ products. I don't buy computer hardware from manufacturers with even a hint or history of misrepresented specs.

 

[groberts101]

problem with all this posturing is that it isn't based on anything more than speculation and a particular users perspective. Most drives(regardless of manufacturer) rarely straight up die and more often than not go into a panic locked/bricked states due to incompatibility, sleep transitions, bad flashes, chipset loading is a big issue for raiders, etc, etc.

The fact that Intel never ventured into the Sandforce territory is what has led them along a much more stable pathway along with much better in-house testing. Had that partnership happened the amount of product R&D alone would have led to improvements for all mfgrs across the board using the controller. Them's some pretty big coattails and you'd be foolish not to ride them if given the opportunity to make more money by doing so.

On a few occasions, I've even seen some guys blaming OCZ for their headaches(and I'm not saying they aren't partially at blame here along with the others riding the Sandforce capitalization wave) only to try 2 other mfgrs using the same controller to ultimately figure out the SF compatibility issues the hard way. SF uses it's vendors to R&D it's controllers and it just so happens that OCZ has sold more of them than some of the other smaller mfgrs combined. This alone can skew the issue(especially for those that like conspiracy theories) and make things appear biased in certain directions.

Personally, I'd like to see the breakdown of sales to RMA requests for each specific controller. Unfortunately, with all OCZ's previous Indilinx RMA's due to panic locks?.. they'd still be bottom of the barrel. lol But then again?.. Indilinx did the very same thing to all it's vendors as well and OCZ sold the P outta those in comparison to others as well. Again with the skewed perspective thing.

All I do know is that OCZ is the ONLY mfgr so far that allows you to request a "special full access destructive flash tool" to recover a panic locked drive to eliminate RMA requests for virtually all of those controllers as of several months ago. No one else(ESPECIALLY Sandforce) wants to allow full firmware access/control with such a tool for fear of custom firmware development and/or reverse engineering.

So until some of you guys know of the specific difference between a "locked" drive and one that has actually "died" due to hardware failure?.. you're just throwin' numbers around.

 

[Voo]

Instead of the marketing literature, search on forum anecdotes, not only here but @ Intel and OCZ. I've seen enough examples of SSD failures attributable to just about every cause.

Interesting, could you link to some examples? Because here on AT (which I think we can agree on is a forum and no marketing literature) I can't remember even one instance of a SSD that exhibited the usual signs of used up PE cycles (can still read just fine, but no writes), but dozens (more hundreds) of locked up/faulty SSDs because of the controller. Although if there were one, it'd surely be a indilinx drive - WAs of 20-35x just are problematic, but we're talking about SF/Intel drives here right?

And we have impartial data that shows that SSDs don't have higher return rates than HDDs (with both 2tb+ 7.2krpm HDDs and Intel SSDs being outliers - only in different directions),

@groberts101: Well still OCZ is responsible for QA (together with SF) and while the root of the problems is obviously SF that's still no carte blanche imho - but one thing we can agree, other SF customers aren't doing a better job at that.

 

[groberts101]

not sure why you felt the urge to write that response when I am fully aware(more than most) when it comes to OZC products. Is why I also included this in the above.

(and I'm not saying they aren't partially at blame here along with the others riding the Sandforce capitalization wave)

If you would have seen some of the crap I've given OCZ over on that forum it would be plain to see the "consumer viewpoint" that I too share with others.