NVIDIA Ion Blu-ray Investigation: Not a problem?

105 Comments

View All Comments

• saiga6360 - Monday, March 16, 2009 - link

  My porn collection says no.

• kilkennycat - Tuesday, March 10, 2009 - link

  Ever seen a storage-cell data-retention-time vs number of write-cycles chart for any SSD? If you have, please post the URL. And have you ever seen a SSD warranty for 5 years?
  
  For a high-tech website, I am highly disappointed in the "follow the crowd" drooling by Anandtech over SSDs. Time for Anand to dig deep into the storage technology of high-density SSD flash-memory with a thorough TECHNICAL analysis of where SSDs can safely be used LONG-TERM and where they should be avoided and hard-disks given preference.
  
  Two essential key components of this analysis will be:-
  
  (a) the extraction of the CELL data-retention-time vs number-of-write-cycles profile from the manufacturers of the flash-storage components in any SSD. Best of luck !!!
  
  (b) the attributes of the wear-leveling algorithms used and how they
  match the intended applications and prior storage profile of the SSD. Plus their impact on access times. (For example, imagine a SSD >85% full of 'static data', the remaining <15% handling rapidly changing data (c.f: Windows virtual memory) and where the "static data" still has to be shuffled to even out the "wear". Moving around that static data every time the dynamic data changes accelerates the overall wear-out... ).
  
  Failing the acquisition of such data, I suggest to Anand the following crude test:- Take a SSD and fill it, say, to 95% with static data. Bang on the remaining 5% with AT LEAST 10 million write-cycles of changing data. That should give the wear leveling algorithms a good workout. At the end of the test, image all the data and store elsewhere. Now leave the SSD statically-powered and check the data-integrity against the stored record every day for at least 3 months. Do not write to the SSD at all at any time during this integrity test. Im sure that the Anandtech crew can come up with a far more vicious test than I. At least, maybe I have prodded the Anandtech elephant to sit up and take notice of the fact that there has not been any technically-critical analysis of data-retention capabilities of the storage elements of SSDs.
  
  Remember, that a SSD with flaky storage due to a single cell with wear-related cell-leakage will have exactly the same symptoms in a PC as flaky memory. And unlike a hard-disk, there is no way to verify sector integrity by a read-write test. (When a hard-disk fails it is generally a one-way street, either a permament sector error or a spin-failure.) A write to the SSD will immediately hide a flaky cell in the wear-leveling process and the SSD will again look perfect, plus depending on the leakage it may take anything between a few days and few weeks for the cell to fail again.... if that cell is not written again in the meantime. Of course, the more times a cell is written, the worse its read-integrity becomes....

• Annirak - Wednesday, March 11, 2009 - link

  [quote]Ever seen a storage-cell data-retention-time vs number of write-cycles chart for any SSD? If you have, please post the URL. And have you ever seen a SSD warranty for 5 years? [/quote]
  You have a reasonably good point. Most SSD's do have a nasty write tolerance issue. Of course, most flash has a nasty write tolerance issue, which is where wear-leveling comes from.
  
  [quote]Remember, that a SSD with flaky storage due to a single cell with wear-related cell-leakage will have exactly the same symptoms in a PC as flaky memory. And unlike a hard-disk, there is no way to verify sector integrity by a read-write test. (When a hard-disk fails it is generally a one-way street, either a permament sector error or a spin-failure.) A write to the SSD will immediately hide a flaky cell in the wear-leveling process and the SSD will again look perfect, plus depending on the leakage it may take anything between a few days and few weeks for the cell to fail again.... if that cell is not written again in the meantime. Of course, the more times a cell is written, the worse its read-integrity becomes.... [/quote]
  You're assuming that the drive has no error checking built in. I don't think that's a good assumption. It's reasonably straightforward to embed CRC on a per-sector basis into a drive like this. That would immediately show up if there were a failure.
  
  While I admit that a lot of the SSD's out there aren't as great as people think, there is one line that appears to be worthwhile. Which is why they cost so much. Anand recognizes the issues with SSDs. Have a look through this page:
  [url=http://www.anandtech.com/cpuchipsets/intel/showdoc...">http://www.anandtech.com/cpuchipsets/intel/showdoc...]Intel X25-M Review[/url]

• kilkennycat - Monday, March 16, 2009 - link

  The Anand article in your reference just swallows Intel's data and extrapolates from that. I expect far more from Anandtech !! When Anandtech takes a SERIOUS look at SSDs and comes up with a suite of INDEPENDENT black-box tests that truly stress the devices in the worst possible way, with the same test patterns being submitted to a reference set of hard-disks, then I might become a believer. The tests would obviously require inclusion of data-latency evaluation under worst-case conditions - for SSDs this would require being at the maximum spec temperature and with power applied, but zero write-data activity --- so the full battery of tests might only be completed after 2-3 months.

• Nacho - Tuesday, March 10, 2009 - link

  What is the point of such a test? You are asking for a test that no normal system will ever reproduce.
  
  All systems write constantly. Even when you are reading a file the system updates the "last access" date, writing the disk.
  
  I don't care if a cell loses it's data in 2 weeks, as long as the wear level algorithm knows this and rewrites the data before that time.

• mindless1 - Tuesday, March 10, 2009 - link

  What would be the point of all that? It's already established that SSD are less failure prone than HDD. It would make more sense to exhaustively test HDD today. If data loss is such a concern, as always a redundant backup is prudent no matter what medium you're storing the primary copy on.

• The0ne - Tuesday, March 10, 2009 - link

  If you're looking for a real technical article, 99% of the tech websites can't give it to you. I don't think many of them are trained or experienced enough to write a technical document properly. Therefore, you can either live with what they are providing in their service or go somewhere else...that 1% in nowhere land.
  
  Seriously, as an engineer, if I wanted tech specs I go get the datasheet and specification documents. If I don't trust what I'm reading I'll review articles on websites such as these or do the tests myself if I have the tools available.
  
  Just my 2cents.

• someonesomewhere - Tuesday, March 10, 2009 - link

  "Do it yourself" is a lame response to someone. If you think more technical specificity is unnecessary, make that case. But, telling someone to become their own review site is rather silly.

• mindless1 - Tuesday, March 10, 2009 - link

  Not so lame, you can't cater anything to everyone, if someone insists they alone need to know something, then they should do the work. If they don't care enough to do it themselves it wasn't really very important was it?

• alphadog - Wednesday, March 11, 2009 - link

  There's a difference between extremes "catering to everyone" and having superficial benchmarks that drive people to erroneously buy SSDs when they shouldn't. Simply putting out exciting benchmarks on R/Ws to empty SSDs is the blind leading the blind.
  
  "If they don't care enough to do it themselves it wasn't really very important was it? "
  
  Or, they don't have the budget, network, and technical access to vendors do properly bench and report on multiple SSDs?
  
  The OP had a very valid point. The bashing just seems like website fanboyism. Some people can't even take constructive criticism...