Excited about Samsung 840 EVO 250GB

[flexy]

Have this SSD coming and this is one piece of h/w I am very excited about, similar as to getting a new graphics card.

(I have an ancient 64GB Supertalent drive as my OS drive which sporadically simply stops working and needs a firmware reflash to get it going again..I absolutely hate this POS Supertalent SSD. Those things had a controller design h/w flaw at least the early models, and its my opinion that they were simply quiet about it but knew about it)

While realizing that the Samsung is probably THE SSD to get today I am somewhat puzzled about the statements about reliability and "endurance" of its memory.

What does this mean? How can a hard drive be "less reliable" or how can "endurance" be an issue? Does it mean that data "all of a sudden" disappears or the drive doesn't read/write correctly anymore unlike "durable" and "more reliable" ones? (I just can't see Samsung or anyone selling drives which would stop working after several months due to "endurance" issues...no sane person would buy such a product).

So what does this really mean in terms of real life use of such a product? Does this mean I can work "reliably" with a MLC drive for a year or two while a TLC based drive would cause problems in the same time?

What does it mean if people say they "would probably not recommend the 840 EVO in a heavy server system"....because it has a high chance of failing or what? Again..if so..how can such h/w even be sold?

 

[Charlie98]

Much the same difference between 'consumer' HDDs and 'enterprise' drives. Build components, warranty and support.

The EVO is a consumer-level component, install it, use it, enjoy it. Don't worry about it.

Having said that, my 840Pro failed in less than a year. Thankfully I have numerous backups so I wasn't hurt by the failure, but the potential is there for ANY component to fail. Make sure you have a viable backup program in place.

 

[Cerb]

So what does this really mean in terms of real life use of such a product? Does this mean I can work "reliably" with a MLC drive for a year or two while a TLC based drive would cause problems in the same time?

No. It means if you plan to run random write-heavy workloads, don't go with TLC. And by write-heavy, I mean like 25GB+/day, mostly in small non-sequential writes, and not allowing the drive too much known free space. IoW, software development and testing with several VMs, OLTP type DBMSs running all the time, or setting the drive up to act a a cache for a RAID array that is too slow with just its spinning platters. Something that you can expect to write TBs to in a short period of time, and/or where WA simply cannot be much lower than ~5x, without tons of OP.

What does it mean if people say they "would probably not recommend the 840 EVO in a heavy server system"....because it has a high chance of failing or what? Again..if so..how can such h/w even be sold?

See above. Tech Report got to ~100 TB before any errors cropped up on their 840 non-Evo, and the drives they used were not allowed to have a great deal of free space, so WA was intended to be a problem (IE, if they filled them to 70%, then let them rest for Windows and the drive to handle TRIM and GC, chances are it would have taken the same drive 2-3x or more the writes to reach that state).

In short, you won't wear it out, but TLC is not fit for some uses, because it wears out faster than MLC.

 

[razel]

FUD about SSD endurance and write performance is OVERBLOWN. Much of it comes from enthusiasts who know little about SSDs writing articles on the Internet. It's a sad fact that it's far too easy to get an audience and make it look like you are an expert. Back in the day a person writing for the paper was a journalist and a guru on the topic. Nowadays with Web 2.0 and reality TV confessional boxes being the normal, one's opinion who happens to be on TV or a website, we now assume as fact.

They just look at the 3000 wear cycles but don't understand that a controller works its butt off spreading the wear over all the NAND on the drive.

If the space is used, the controller makes decisions on how to deal with it. With extra space around 25% it can keep it's WRITE performance high in VERY demanding server constant write situations like what Anand's been showing. Outside of that, in real life you'll never or rarely use data on your computer at that rate.

So let's make life even worse for the SSD, if it the drive was near full and it want's to update a file, of course it will shuffle data around to even out the wear. It will want to write to a NAND block that's written to least. Since it has to shuffle data around, that's where writes will slow down, but it still will be MILES faster than an HDD. In fact, in real life, doesn't that slowness happen with full HDD space too but for other reasons? Also with a full drive SSD or HDD, you got that problem to worry about!

So ignore the FUD. Endurance and performance is not an issue for 95% of people. Even performance differences are a bit overblown too. You can bench the difference but in real everyday usage it may not be noticeable 95%. If you did notice it, the price difference won't be worth it. So just buy any current model SSD and worry about size, price, and warranty.

 

[Morbus]

It's only noticeable in the writes, and even then, only when they're consistent and you know what to expect. And even then it won't be that big a difference that you might think something's broken. Just about 30% slower for most people on a full SSD.

Over-provision solves most of those issues though.

As for the write wear or whatever, people really have absolutely no idea how little they write to their hard drives. I have a couple of games, Adobe CS and a couple of other programs on my SSD. I started using it a month ago (and I'm at my computer all day, so...), and just checked my writes: 0.56TB. And that's counting the installation of all the software so the average will be much lower. But eve if it was 500GB a month, that averages 6TB a year. In ten years time, maybe I'll get some reallocated sectors on SMART. If I'm unlucky...

Well, if I'm really unlucky, the SSD's controller will die in a few months, but that kind of failures will happen in any software. I think that if lasts the two year warranty, I should be set for at least five more years after than. And before the drive wears out or dies, I'll be replacing it anyway (because that's the way of things)...

My real fear is that the drive I'll be replacing it with will be even more fragile and will last even less... Because that's the way things are heading towards, unfortunately...

 

[Cerb]

I've been doing 30GB+/day on my M500, lately (it doesn't have smart values to check, so it could be much more in NAND writes, plus whatever writing Windows is doing day to day), just using folder copy to do Skyrim backups, to track down my CTDs, without giving up added functionality and content, where possible (fresh install, ini mod, add mods til it breaks, start turning some off, changing to others, clean masters, etc.). If I kept that up (not likely), and were actually doing 50GB of NAND writes/day (probably close, w/ no compression, and big pages, but also probably a bit higher than actual), it would take a week and half to perform one write over the drive.

That leads to needing about 75 years of doing that to reach the flash's p/e rating, at which point the flash still will not have failed, plus it has additional error correction built in, compared to many other SSDs. The controller chip, DRAM chip, or capacitors might wear out...Hell, the SATA contacts might corrode, and the solder balls crack, decades before the flash wears out. It won't last until the flash wears out, I guarantee you.

Now, to 2009, drives with 50x WA, and some with 100x+ WA, existed, and weren't uncommon, in terms of what was available, between flash with huge pages, controllers writing 512B at a time, and controllers not being good at allocating writes to minimize movement of static data. Somebody here,some months ago, showed the write stats from a Patriot Torqx, FI, showing that it was getting >150x WA in regular use. If you weren't paying attention ca. 2010, Intel's X2, offering good performance with low write amplification, was a really big deal.

Up until the shrink to ~20nm, the improvements in ECC and wear leveling were so thoroughly improved that a new MLC drive, averaging <3x WA for most users (some drives with some users well under 1.5x, but power users and content creation people seem to be able to report much higher), is likely to outlast an old SLC drive. But, now, they're about at the end of what they can do, so endurance reduction with shrinks may turn into a big deal.

But, not today. Today, if you have to ask, the answer is probably no, you don't have to care. And, if you do have to care, it's just a matter of planning out your costs, because you're either looking at disks that are too slow, or DRAM that is too expensive, and it's just a question of what the best implementation type, and then series/models. If you where them out over time, they will have still paid for themselves, because in many cases, no spinning disk array is capable of matching a single SSD, and SSD caches split the difference with big storage just like they do on regular PCs (business-market SSDs capable of sustaining 5000 or more random IOPS, while also still maintaining <10ms access time with full queues are widely available, yet there is no spinning array in the world that can do the same).

 

[rovopio]

how does an ssd fail? is it just not being able to write on it anymore... or is it fail like hdd fail, like not recognized and data gone?

i bought an intel ssd to be put on my laptop and it has survived 3 laptops until now. 2 laptops died before the warranty runs out.

 

[Cerb]

how does an ssd fail? is it just not being able to write on it anymore... or is it fail like hdd fail, like not recognized and data gone?

It varies. The SSD is a whole little tiny computer unto itself, with the SATA port as a networking interface to the outside world. The controller has one ore more generic CPU cores, typically, the SATA interface, flash translation layer hardware, some magic special hardware accelerators, and physical interfaces to several channels of bare flash. Most of the rest of the magic is in software written on it.

Bad pages or blocks happen, even on nominally healthy NAND.

Firmwares have bugs.

Controllers have bugs.

Or, random **** happens.

On the whole, most are pretty reliable, but HDDs are sufficiently mature that there's only a few things that can go bad, outside of occasional latent firmware bugs.

 

[Morbus]

Or people banging the laptops around and letting the HDD fall... SSDs are better for that, but they are not indestructible.

 

[Hdgamer]

I've had two out of my three ssd's fail on me so far and none of them due to the endurance issues. Controllers fail, the electronics on it will take a crap eventually. Both of these were MLC older gen ssd's. One was a gen 2 Intel 80 gig, and a 120 gig Vertex 3. Both died the same way, OS was stuttering, bios wouldn't recognize it sometimes, then all of a sudden dead. As far as I'm concerned, I would be happy with three years out of an OS drive. My hdd's so far have been phenominal. I've owned 20 hdd's in the last 15 years and only 2 died after a nice long life. Mainly the death was my fault too.