ADATA Releases Three SSDs - Maximizing The Capacity of SandForce Drives

32 Comments

View All Comments

• jwilliams4200 - Saturday, February 25, 2012 - link

  This article is misleading.
  
  Most Sandforce drives have flash set aside for both overprovisioning (OP) and redundant array of independent silicon elements (RAISE).
  
  For example, take a SF2281 SSD with 128GiB of flash on the circuit board (usually advertised as "120GB SSD"). Typically, 8GiB of the flash is used for RAISE. That leaves 120GiB of flash for storage and OP.
  
  Since the SSD is advertised as "120GB", it must be capable of storing at least 120,000,000,000 bytes. Since there is still 120GiB = 120*(1024^3) = 128,849,018,880 bytes of flash left after subtracting 8GiB for RAISE, that leaves 128,849,018,880 - 120,000,000,000 = 8,849,018,880 bytes available for OP.
  
  How much OP is that? 128,849,018,880 / 120,000,000,000 = (1024^3) / (1000^3) = 7.37%
  
  In other words, most "120GB" Sandforce SSDs have both 7.37% OP and 8GiB of flash used for RAISE.
  
  But this Adata drive advertises 128GB of storage capacity. If that is true (and not just a misrepresentation like OCZ did last year), then Adata must have dropped either OP or RAISE.
  
  I find it very unlikely that they would drop all overprovisioning, since that would seriously hurt performance and longevity.
  
  So, most likely what they have done is dropped RAISE. That is not especially surprising, since Intel also dropped RAISE in their 520 SSD.
  
  But this misleading article makes no mention of RAISE at all. And cites a source, thessdreview, which gets this completely wrong, talking about "0% OP".
  
  I am disappointed in the quality of the journalism here from anandtech. I would have expected better.

• jwilliams4200 - Saturday, February 25, 2012 - link

  By the way, for those who are not familiar with RAISE, it is similar to RAID, in that a parity block is set aside for each X number of data blocks.
  
  For example, with a typical Sandforce SSD with 128GiB of flash on board, 8GiB of flash will be devoted to RAISE. That means that there are 15 data blocks and 1 parity block (similar to a RAID 5 with 15 data disks and 1 parity disk).
  
  The reason this article is misleading is because the overprovisioning (OP) and the RAISE in a typical "120GB" Sandforce SSD are each given approximately 8GiB of flash (to be more accurate, 8GiB for RAISE, 8.24GiB for OP).
  
  In the case of Adata, they have increased the usable storage capacity from 120GB to 128GB, meaning they must have dropped either RAISE or OP to get that extra ~8GB usable.
  
  Since it is extremely unlikely that they dropped the overprovisioning, the conclusion is that they dropped the RAISE. This makes sense, because RAISE only functions to increase the reliability of the SSD if a large block of flash dies, so dropping RAISE only decreases the reliability of the SSD. Eliminating overprovisioning, in contrast, would drastically reduce both the performance and the longevity of the SSD, and no rational designer would do such a thing.
  
  It may be that the author of this article has confused overprovisioning with RAISE, and that is why he has written such a misleading article.
  
  Bottom line is that the Adata SSDs probably have the same overprovisioning as other SSDs (about 7%) but that they have dropped RAISE and so are not protected from failure when a large chunk of flash memory fails.

• JarredWalton - Saturday, February 25, 2012 - link

  Or ADATA may have decreased RAISE and OP to ~4% instead of 7-8%. They haven't exactly stated what they are doing, merely that their drives are now 128GB instead of 120GB.

• jwilliams4200 - Saturday, February 25, 2012 - link

  No, that is not supported by the SF2281 controller. To use RAISE, you must devote at least one whole die to RAISE, which would be 8GiB in most cases.

• JarredWalton - Saturday, February 25, 2012 - link

  True. I suppose they could either eliminate RAISE or eliminate OP, and I'm with you in guessing they sacrificed potential reliability (RAISE) rather than OP. If they kept RAISE and ditched OP, they'd actually have a 129GB (128.8GB) SSD. :p

• Kristian Vättö - Saturday, February 25, 2012 - link

  Intel 520 Series is still sold as a 120GB drive, though, and has the same usable capacity as OCZ Vertex 3 (which has RAISE enabled). Shouldn't Intel be a 128GB drive then, or does Intel use more space for OP? This isn't making much sense now.
  
  To be honest, I had no reason doubt TheSSDReview's explanation because it was logical, especially because they had been in contact with LSI (aka SandForce). If it had been their own analysis, then I would have taken it with a grain of salt and most likely confirm it before posting this. I think they are correct because disabling RAISE didn't turn Intel 520 Series into a 128GB drive, so "0% OP" would explain that.
  
  I have emailed LSI and ADATA to ask for some clarification but it may take a few days now that it's weekend. We'll see...

• jwilliams4200 - Saturday, February 25, 2012 - link

  I don't know for certain that Intel disabled RAISE on the 520, but there is one persuasive piece of evidence that makes me think that they did:
  
  The UBER spec on the Intel 520 is 10^-16.
  
  http://www.intel.com/content/www/us/en/solid-state...
  
  However, Sandforce specifies that RAISE will improve the UBER up to 10^-29
  
  http://www.sandforce.com/index.php?id=174&pare...
  
  I think Intel disabled RAISE on the 520 and increased the overprovisioning in order to increase performance and longevity. If I am correct, the 120GB Intel 520 uses about 14% OP and no RAISE.

• Kristian Vättö - Sunday, February 26, 2012 - link

  Tom's Hardware is also saying that RAISE is disabled on 520 Series
  
  http://www.tomshardware.com/reviews/ssd-520-sandfo...
  
  I also asked this on the email so hopefully we will get a clarification.
  
  I still think there must be something with the OP as well. Most 60GB drives have RAISE disabled because they consist of only eight 8GiB dies. Okay, there is just enough space to market the drive as 60GB if 8GiB is used for RAISE but it doesn't sound right that only ~130MB would be used for OP, especially because we are talking about the old firmware which requires ~7% OP. Hence I think something must have been done to the OP in order for ADATA to achieve a usable capacity of 64GB.

• Kamil_FPC - Saturday, February 25, 2012 - link

  Intel didn't dropped RAISE, why do you think they do this? SSD 520 is a standard SandForce based drive, with ~14% of OP, some of it is probably used to support RAISE.
  
  I don't know the details behind AData drives and new LSI SandForce firmware, but I suspect it still supports RAISE since it is one of core function in this controller, but they are not using the whole NAND, they don't have to. The difference betwen GB and GiB is enough for OP and RAISE.
  
  We will know for sure when Anand put his hands on AData products :)

• jwilliams4200 - Saturday, February 25, 2012 - link

  The SF2281 controller has some restrictions on RAISE, specifically, it requires at least one die of flash be devoted to RAISE. Since SSDs with 128GiB on the circuit board generally use 8GiB or larger dies, they must devote no less than 8GiB to RAISE.
  
  So it is unlikely that Adata is using 4GiB for RAISE and 4GiB for OP in their "128GB" SSD.