SSD Reliability these days...

[hhhd1]

Intel apparently mislead the press because even we thought that Intel had source code access but LSI/SandForce informed us later that even Intel didn't get full access to the source code.

Is it possible to know how much control does intel have over the sandforce firmware ?

It seems that they have their unique issues, for example, there are some issues working with AMD motherboards, and also, the newest model (intel 530 ssd) having issues being recognized by the system after a soft reboot. ... Those issues that are unique to intel, which suggest that they have some sort of special customization ?

Intel soft restart issue: https://communities.intel.com/thread/44258

 

[GlacierFreeze]

Oh, and put the page file on the SSD. Everything wears out the SSD faster that writes to it. That it writes really fast is the point. Unless you're running Windows in 1GB RAM, you're not going to bring your SSD to an early death by having the PF on. It takes much worse than that to kill them early.

Hmm, I'd have to disagree. I'd definitely recommend putting PageFile on a separate, mechanical drive.

http://forums.anandtech.com/showthread.php?t=2354701

Minor update to that. It's been on 0.73 TB written since I did that, as opposed to +0.01 TB everyday. No telling what other games stream writes to PF. Probably not many at all, but I'd rather be safe than sorry.

 

[Charlie98]

I game on mine they are good enough now a days to be regular use drives.

That's what I do, part of the benefit of an SSD is fast loading of levels, terrain and such. I got a big enough SSD that I can fit everything on the computer onto it, including all my game files.

I have to scratch my head... the SSD is a tool, a part.... a consumable part. Yes, eventually it gets 'used up,' for lack of a better term, and you have to replace it. But I have never seen the value in buying a high-performance part and then babying it, it's like buying a set of tires and not going over 30MPH because you will 'use them up' faster at 60MPH. We're not talking $1000 here (at least not consumer-grade SSDs; ) by the time my Plextor 256GB drive needs replacement, there will something better and probably cheaper available... it's all good. 😀

 

[John Connor]

Yep! I have the same philosophy, especially about loading levels in games.

 

[Cerb]

Hmm, I'd have to disagree. I'd definitely recommend putting PageFile on a separate, mechanical drive.

http://forums.anandtech.com/showthread.php?t=2354701

Minor update to that. It's been on 0.73 TB written since I did that, as opposed to +0.01 TB everyday. No telling what other games stream writes to PF. Probably not many at all, but I'd rather be safe than sorry.

Even .01TB every day is only 3.65TB every year. Even small TLC drives will run out of their warranty time period long before that makes for a problem, even if the WA from that 10GB is fairly high.

 

[Emulex]

I put the whole esxi server on the samsung 840 pro. the old 830's are probably at 10x past their MWI=0 limit and holding steady.

The key to longevity is overprovisioning, small blocksize (8k page/strip size), raid-1 for redundancy if you care.

Do I give a carp if a drive fails? No! To afford the 50 SSD drives worth of raid-1 storage, I would have had 400 sas drives to give equivalent IOPS. These sammy drives are bullet proof, you just need to know how to treat them.

Consumable is the key word - everything is consumable. Nothing lasts forever.

Charlie98 has it dead on - drive it like you stole it, and enjoy!

 

[dnut_00]

That's what I do, part of the benefit of an SSD is fast loading of levels, terrain and such. I got a big enough SSD that I can fit everything on the computer onto it, including all my game files.

I have to scratch my head... the SSD is a tool, a part.... a consumable part. Yes, eventually it gets 'used up,' for lack of a better term, and you have to replace it. But I have never seen the value in buying a high-performance part and then babying it, it's like buying a set of tires and not going over 30MPH because you will 'use them up' faster at 60MPH. We're not talking $1000 here (at least not consumer-grade SSDs; ) by the time my Plextor 256GB drive needs replacement, there will something better and probably cheaper available... it's all good. 😀

That's not the correct allegory. The correct one is to get a set of new tires and do whatever on nice interstates, but get your speed below 30MPH on rough terrain.

 

[Morbus]

I'm running Windows 7 on a Samsung 840 EVO 250GB, and I have my page file on a separate HDD. I have noticed absolutely no performance hit whatsoever, the HDD goes to sleep just as normally, it's almost as if the page file isn't being used...

So, from my experience, get that page file off the SSD, you won't be any worse off, and you save some valuable space (I do, at least, since I'm running 16GB RAM and having a 20GB page file isn't really all that great) which you can use for over provisioning.

As for games running off the SSD, newer games will benefit more than older games because of the larger sizes of their files. Remember that the biggest performance boost in load times is RAM and CPU, because the games are actually computing the level data when loading, not just read it off the drives... That's the reason a good PC will load games faster, despite having pretty much the same drive as a 10yo PC.

But having games on an SSD shouldn't affect its life significantly. Reading off a SSD isn't the problem, writing is.

That, or having a bad SSD. From what I've read around, plenty of brands have very high failure rates, and, at least for me, any failure rate is too high. I expect my drives to last for as long as it's physically possible, and not to die from some controller malfunction or some crap like that.

 

[mikeyanrol]

The Samsungs are pretty well proven, not so much the EVO just yet, but the Pro. Having said that, my 840Pro died on me last month with not a year of service. Luck of the draw... I wouldn't hesitate to buy another one. After the 840Pro died, I grabbed a Plextor M5Pro as a replacement, it has been solid all the way, but the Plextor has a horrible toolbox, if that's of importance to you. I would add the Plextor M5Pro to the lineup above.

I just gathered parts for a general build for my inlaws, I wanted it as foolproof and reliable as I could get it... I picked an Intel 530, even considering the SandForce controller.

I also have a OCZ Agility3 that's going on 2 years old... never had a lick of problems with it, either. OCZ had some real QC issues in their 2nd and 3rd generation SSDs, I think mostly linked to the SandForce controller and their FW for it, but it generated a lot of hate for the OCZ brand.... not undeserved. I would not buy another OCZ drive for a number of reasons even though I have gotten very good service out of what is a lower performing drive (the Agility line) and the fact they are selling at giveaway prices.

A lot of people fret over the amount of writes to an SSD and effect on it's life, I don't worry about it... install it and use it. It's a consumable part. Monitor it with the toolbox (if provided) and CrystalDisk and be happy. I would also back up the OS drive just in case it dies... it happens (obviously.) I do daily full version backups of my OS SSD every day using Acronis... when my Samsung puked, I installed the Plextor, mounted my backup image and I was back in business. The build for my inlaws, even though it will have an Intel SSD... will also have Acronis and be backed up every week or so.

I believe most of the issues currenly the users are having across these OCZ ssds will surely get resolved if they get the latest firmware updates.

Thanks.

What's so horrible about a Sandforce Controller? What dod it do or not do that it should?

Yes SF had some reliability issues in the past, but guess its all resolved now.
You may have read a lot about SandForce Controller issues, but that is a thing of past. The way to check it is with the sequential benchmarks for the SSD to it's maximum with programs that issue I/O operations at a much faster rate than an application can, and does so at high queue levels....

 

[HeXen]

I have several and most of them are several years old. I still have one of the early single cell ssd's when they first came to market. Reliable isn't the word for how much abuse they can take. I don't think the word exists.

 

[sub.mesa]

Yes SF had some reliability issues in the past, but guess its all resolved now.

The issues can never be resolved because they are - in part - inherent to the design of the Sandforce controller. Due to the deduplication functionality, this SSD controller will always have an additional point of failure others SSDs don't have. Other SSDs only have a weakness in the unprotected mapping tables, while Sandforce also has to protect the integrity and consistency of its deduplication tables. Hence, at least on paper it will never achieve the reliability of other SSDs.

That is not to say Sandforce SSDs cannot be reliable. But to say the issues are 'resolved' is probably an incorrect assessment. Due to the additional complexity, a lot of bugs popped up that took years to iron out. Now that is happened, SSDs sporting this controller don't have shockingly high failure rates any longer, such as in the past.

Besides, any NAND storage device without capacitor protection is unreliable by design.

 

[Hellhammer]

The issues can never be resolved because they are - in part - inherent to the design of the Sandforce controller. Due to the deduplication functionality, this SSD controller will always have an additional point of failure others SSDs don't have. Other SSDs only have a weakness in the unprotected mapping tables, while Sandforce also has to protect the integrity and consistency of its deduplication tables. Hence, at least on paper it will never achieve the reliability of other SSDs.

Actually, SandForce SSDs are more reliable in that sense because they don't use DRAM to cache the mapping table. Instead they rely on the controller's internal SRAM, which is still volatile by nature but the data isn't fully lost after the power is cut. Furthermore, I think SF's mapping table design is different from others because the SRAM can't hold the whole table (IIRC SF-2281 has 8MB of SRAM cache), in which case most of the table is stored in NAND.

 

[sub.mesa]

Actually, SandForce SSDs are more reliable in that sense because they don't use DRAM to cache the mapping table. Instead they rely on the controller's internal SRAM, which is still volatile by nature but the data isn't fully lost after the power is cut. Furthermore, I think SF's mapping table design is different from others because the SRAM can't hold the whole table (IIRC SF-2281 has 8MB of SRAM cache), in which case most of the table is stored in NAND.

Are you sure that is correct? As far as I recall, the internal SRAM is used for write-back, while external DRAM is (optionally) used for caching of the mapping tables. I could be wrong on Sandforce though, but at least this applies to other controllers like Intel X25-M, 320, S3500/S3700.

Caching of mapping tables isn't dangerous - using DRAM as write-back is. Samsung falls in the category where user data goes through the DRAM. Crucial M500 as well, but only a few megabytes and protected by cache flush commands.

I would dispute that Sandforce is more reliable than its direct competitors. But the extremely high failure rates like in the Vertex 2 days are indeed past tense.

 

[Hellhammer]

Are you sure that is correct? As far as I recall, the internal SRAM is used for write-back, while external DRAM is (optionally) used for caching of the mapping tables. I could be wrong on Sandforce though, but at least this applies to other controllers like Intel X25-M, 320, S3500/S3700.

SandForce has never required or even supported external DRAM, it's one of their unique features. I know the SRAM is used to cache the mapping table but I'm not sure if it's used as a write-back as well (LSI/SandForce has never been very open about the details of their architecture).

However, note that all SSDs use some sort of cache as a write-back due to write combining. Without any user data caching the write amplification would jump through the roof since IOs smaller than the page size would still be written to one page (e.g. 4KB IO would have to be written as 8KB).

Caching of mapping tables isn't dangerous - using DRAM as write-back is. Samsung falls in the category where user data goes through the DRAM. Crucial M500 as well, but only a few megabytes and protected by cache flush commands.

I wouldn't say so. Caching user data may lead to a corrupted file if the data is lost but a corrupted mapping table means that the SSD just lost track of all your data.

By the way, do you have any source for the usage of DRAM for user data? Interested in seeing because it's not something we can easily test and OEMs don't usually like to talk about that.

I would dispute that Sandforce is more reliable than its direct competitors. But the extremely high failure rates like in the Vertex 2 days are indeed past tense.

I didn't mean that SandForce as a whole is more reliable, just that from that angle it is.

 

[Ao1]

By the way, do you have any source for the usage of DRAM for user data? Interested in seeing because it's not something we can easily test and OEMs don't usually like to talk about that.

If disabling “Enable write cache” makes no difference to performance that is one clue. Intense random 4K write activity will also make the DRAM get very hot, which is another clue.