SandForce-based SSDs have lower-quality NAND?

[VirtualLarry]

By writing less to flash SandForce also believes its controllers allow SSD makers to use lower grade flash. Most MLC NAND flash on the market today is built for USB sticks or CF/SD cards. These applications have very minimal write cycle requirements. Toss some of this flash into an SSD and you’ll eventually start losing data.

Intel and other top tier SSD makers tackle this issue by using only the highest grade NAND available on the market. They take it seriously because most users don’t back up and losing your primary drive, especially when it’s supposed to be on more reliable storage, can be catastrophic.

http://www.anandtech.com/show/2899/4

 

[taltamir]

keyword "allows". SandForce is saying that they believe their controller is so much better than others, that you could safely use low grade flash on it. However, AFAIK nobody does... Also AFAIK, they aren't the only controller that COULD use lower grade MLC.

finally, less writes is really not as big a deal anymore. most people will take hundreds of years to use it up.

 

[Mark R]

Sandforce SSDs use the same grade flash as other models of MLC SSD.

The difference is that Sandforce have gone out of their way to make their controller the most reliable controller on the Market, and the one that gets the most use out of the flash chips.

Extreme budget flash targeted at USB sticks is unlikely to be suitable for SSDs as it is too slow, and it's cycle life time is hugely worse than normal consumer grade MLC.

What Sandforce have aimed to do with their advanced controllers is to get SLC reliability from MLC flash. It's not good business to sell a premium controller with garbage grade flash, giving an average product. The Sandforce chips are aimed at enterprise, but allow the use of consumer grade MLC instead of pricey SLC.

Of course, just because the features are available doesn't mean they'll get used - RAISE ends up turned off on a number of consumer level SSDs because it sacrifices some space.

 

[sub.mesa]

3-bit MLC is only used in CompactFlash and USB sticks. As far as i know, no SSDs are using it. I wouldn't advise buying one, either.

For USB pendrive/CF 3-bit MLC can be just fine; to store large files like pictures on your camera which never get overwritten. Low write amplification, 3bit MLC would be about half as cheap to make as 2bit MLC for the same capacity.

 

[Emulex]

just like ram/cpu's they bin the chips - they will never use 3 bit for mlc with current technology. enterprise mlc is going to demand the best 2 bit mlc.

like a 2TB drive - each one has bad sectors from the get go - and so much spare sectors. the good ones become RE4/AV/etc the junkier ones go to blue/black and the worst go to externals.

 

[adamantinepiggy]

Remember too that Sandforce is not a memory maker, nor are the vast majority (if not all) of the SSD makers who use their controller. Unlike Intel, Micron, Samsung, etc, being NAND memory "makers" who get to cherry pick and have internal supplies of NAND memory to use how they wish at cost. the makers of the Sandforce SSD's have to purchase what's available at a reasonable price on the open chip markets. Since supplies of top tier open-market NAND might become prohibitively expensive, it's good to hedge your bets by being able to work with lower quality NAND.

This is one of the reasons, you most likely wont see large scale adoption of the Sandforce controller SSD's in the big brands like Apple, HP, Dell, etc as an OEM-branded drive options. For the large companies, it's much safer to go with a brand where the most volatile priced parts (NAND memory pricing for example) are all made and assembled by a single manufacturer as the needed volumes for this market are huge. The big brands also "demand" custom firmware on set model-designs that would be difficult for a company like Sandforce to produce since they don't manufacture the end-product SSD's their controller go into and have limited control on the final end-hardware.

Even if these controllers are the best on the market, possible market volatility on other component prices keeps all the small non-memory maker SSD's out of the most lucrative large OEM branding contracts.

 

[taltamir]

Remember too that Sandforce is not a memory maker, nor are the vast majority (if not all) of the SSD makers who use their controller. Unlike Intel, Micron, Samsung, etc, being NAND memory "makers"

Who is that etc?

 

[Mark R]

Remember too that Sandforce is not a memory maker, nor are the vast majority (if not all) of the SSD makers who use their controller. Unlike Intel, Micron, Samsung, etc, being NAND memory "makers" who get to cherry pick and have internal supplies of NAND memory to use how they wish at cost. the makers of the Sandforce SSD's have to purchase what's available at a reasonable price on the open chip markets. Since supplies of top tier open-market NAND might become prohibitively expensive, it's good to hedge your bets by being able to work with lower quality NAND.

...

The big brands also "demand" custom firmware on set model-designs that would be difficult for a company like Sandforce to produce since they don't manufacture the end-product SSD's their controller go into and have limited control on the final end-hardware.

Why would a flash manufacturer want to 'cherry pick' when they can sell the best flash at a substantial premium as enterprise MLC? eMLC is flash that is specified for 20-30k write cycles and better data integrity - it's the exact same dies as consumer-grade MLC, but it's the top bins (the eMLC chips are underclocked too).

The other thing is that Sandforce is the one of only 2 companies that have actually tested multiple types of flash, and developed firmware specifically tuned to the requirements of specific flash chips (the other company is fusionIO). Every other SSD controller treats the connected flash as generic flash without brand specific optimizations. This means that an OEM buying sandforce controllers can source multiple types of flash memory, and can be sure that the controller is fully-optimized for all types of flash that they may source.

 

[VirtualLarry]

Of course, just because the features are available doesn't mean they'll get used - RAISE ends up turned off on a number of consumer level SSDs because it sacrifices some space.

ok, now THAT disturbs me. that's one of sandforces most highly touted features. why would oems disable that. if you can tell me which drives have it disabled, I would like to know.

I just ordered an a-data s599 64gb model.

 

[LokutusofBorg]

I just ordered an a-data s599 64gb model.

The ADATA one definitely disables it.

I'm just kidding. You were so excited to order that in the other thread, I couldn't resist. 😉

 

[capeconsultant]

Darn, this is all wicked interesting stuff. For instance, I always wondered if external WD drives were the precise equivalent of internal ones. Here you say they are bottom of the BIN. OUCH. I recommend these drive to clients for backup.

All this info about who uses which NAND with which controller with which features turned on/off is highly interesting.

I like any company that pays attention to quality. Sandforce now has my attention. Sure would like to get my hands on a 2000 🙂

Thanks for all the great info, keep it coming! I love SSD's and will never ever ever go back. Mr. Jobs sees the future correctly regarding SSD's. Yes he does!

 

[Mark R]

ok, now THAT disturbs me. that's one of sandforces most highly touted features. why would oems disable that. if you can tell me which drives have it disabled, I would like to know.

I just ordered an a-data s599 64gb model.

It comes down to the 'missing' capacity. RAISE works like RAID5/6 - it sacrifices capacity of a flash die, in order to use it as parity.

The Sf controller has 8 channels, so dedicating one flash chip to parity wil cost 12.5% of available space. Why do you think the first Sf drives shipped with 28% 'missing' capacity - 64 GiB of flash but only 50 GB?

Theoreticlly, some flash chips are multi-die modules - and RAISE is smart enough to be able to Run RAISE across the physical dies, rather than the actual chips. However, 8GiB dies are pretty standard stuff at 34 nm, so I doubt that stacked modules are used much at this capacity point. Can't be sure though- the manufacturers treat their flash specs like top secret material, so you can't find out for love nor money.

Still, a 64GB drive with 64GiB of flash has only 7% hidden. This is too small for a useful level of RAISE, even if 2-stacked flash was used (6.3% for raise - 0.7% for alll other SSD functions).

By contrast fusionIo have 25 channels, dedicating 3 for parity in RAISE-Z3. This eats 12% of the space in one go.

 

[Dark Shroud]

What it seems to come down to this next round is do you speed or stability. As long as the Sandforce drives use Samsung NAND you should have a solid drive. Meanwhile Intel/Micron have the best NAND but Intel isn't doing SATA 6gps while Micron is for some models.

Personally I'll take the Intel 300GB or 600GB drives over faster Sandforce drives.

 

[taltamir]

What it seems to come down to this next round is do you speed or stability. As long as the Sandforce drives use Samsung NAND you should have a solid drive. Meanwhile Intel/Micron have the best NAND but Intel isn't doing SATA 6gps while Micron is for some models.

Personally I'll take the Intel 300GB or 600GB drives over faster Sandforce drives.

the sandforce drives are not going to be less "stable" (reliable?)

 

[cliftonite]

Who is that etc?

Toshiba?

 

[adamantinepiggy]

Why would a flash manufacturer want to 'cherry pick' when they can sell the best flash at a substantial premium as enterprise MLC? eMLC is flash that is specified for 20-30k write cycles and better data integrity - it's the exact same dies as consumer-grade MLC, but it's the top bins (the eMLC chips are underclocked too).

By cherry picking, it's more about being able to reserve "quantity" of top-tier NAND, not so much particular quality within the same series (like OC'able NVidia GPU binning). Unless the SSD divisions within the memory-makers suck, the company should "normally" be making quite a bit more for for each NAND chip sold on a SSD versus selling the same chip on the open market. It would be pointless for a memory-maker to continue internal SSD research/manufacturing/sales if they are losing money on any long-term component sales.

The other thing is that Sandforce is the one of only 2 companies that have actually tested multiple types of flash, and developed firmware specifically tuned to the requirements of specific flash chips (the other company is fusionIO). Every other SSD controller treats the connected flash as generic flash without brand specific optimizations. This means that an OEM buying sandforce controllers can source multiple types of flash memory, and can be sure that the controller is fully-optimized for all types of flash that they may source.

As the memory-maker companies are also generally their sole-source supply of NAND chips used in their own SSD's, why exactly do they need to test other types of flash not made by the company? They are not dependent on open-market pricing for determination of what memory chips they can use.

In addition, the OEM request-for-bid contracts "stipulate" what memory, length of production, board design, etc the product they want "must" have for the life of the product. So for XX months/years part of the manufacturer MUST be able to supply the "exact" same parts regardless of open-market pricing volatility, regardless if new/better chip design comes about. So for non-memory-maker SSD producers, they cannot easily enter these types of agreements as they do not have control of what NAND is available for them.

For example, the Apple-required contractual production stipulations of the SSD in the new Apple Mac Air are EXTREMELY stringent I know because our company lost out because we were in the middle of a NAND series change; basically, we couldn't provide long-term guarantee of our older product manufacturing, but our newer product was not fully ready in supply/quantity ability.

How the SSD is made, what exact memory chip parts it is made up of, and etc for this month, must be the exact same until the product's end-of-life cycle for these OEM-BRANDED type products. Unless the SSD company makes the majority of its own parts (at least the most volatile price/quantity parts like NAND), it is near impossible to even attempt to bid for these type of contracts. Even if a better/cheaper NAND chip design/series comes up for the manufacturer, it doesn't matter, unless allowed by the other party (in this case Apple) and they are very unlikely to allow any "hardware" changes beyond engineering fixes.

For a non-memory-maker company like OCZ, Kingston, or whomever use the Sandforce controller etc and who have no guaranteed long-term supplies of the exact same NAND or other compnents, it's near impossible to get these OEM-branded contracts. It's hard enough when you make your own stuff.

 

[Emulex]

^^ this is why samsung and toshiba for the most part are oem-only. they cater to this market. while not the fastest - their product is consistently quality and the same. you can't deploy 1 million laptops and have 20% of them flake out due to a change in production. that would not make anyone happy

 

[taltamir]

Toshiba?

thanks for clarifying, does anyone actually use toshiba chips in SSD? AFAIK only intel-micron (intel and micron co-own a subsidiary that produces the chips for both of them together) and samsung chips are used