back to article SanDisk Extreme Pro SSD – courting speed freaks and gamers

Typical. Just when we've gotten over the shock of a SSD drive being offered with a 10-year warranty, namely Samsung’s SSD850 PRO (reviewed recently), when another appears on the block, this time from SanDisk. SanDisk Extreme Pro SSD SanDisk Extreme Pro SSD I wonder if you asked the man in the street to name his top four …

  1. Alan Brown Silver badge

    write latency scatter?

    As important as write speeds, is write latency. It's no good having high throughput if the drive hiccups occasionally.

    Sandisk have a paper up about this at http://www.sandisk.com/assets/docs/performance_consistency_final_08_26_13.pdf

    Several websites started looking at this seriously when intel pointed it out as part of their datacentre range release - and sure enough a lot of consumer-grade drives sometimes take tens or hundreds of milliseconds to complete a random write or read. This has a huge impact on overall performance.

    A good example of the differences between drives is shown in the charts at http://www.storagereview.com/intel_ssd_dc_s3700_series_enterprise_ssd_review

    It'd be extremely interesting to see the scatter plots for the 850pro and the Extreme Pro - this is the kind of thing which makes the difference between "fast" and "not as good as it should be"

    1. Sandtitz Silver badge

      Re: write latency scatter?

      It'd be extremely interesting to see the scatter plots for the 850pro and the Extreme Pro

      Go to Anandtech then to read the graphs.

      They started testing the performance consistency as part of their SSD reviews after Intel lauded that aspect with S3700.

  2. Anonymous Coward
    Anonymous Coward

    The price points are getting interesting now the device life is more than a few years.

    Disk drive prices have become so low that it is easy to forget what you get for the money on this SSD. As a reality check - in the early 1990s a PC 500MB disk would cost about £175. Circa 1982 - the add-on 10MB hard disk for an Apple II was the best part of £2K. All prices without allowing for inflation. It is possible that an Apple II 5.25 floppy drive cost nearly as much as one of these SSDs.

    1. Jim 59

      Can't find the joke icon

      Circa 1982 - the add-on 10MB hard disk for an Apple II was the best part of £2K.

      2K in 1982 = about 6k in 2014. That is quite cheap for an Apple peripheral.

    2. Alan Brown Silver badge

      "In the early 1990s a PC 500MB disk would cost about £175."

      In 1991 a 200Mb drive cost me US$1200

      In 1993 the same money got 1Gb

      in 1995, 4Gb and 7200rpm (barracudas, which ran so hot that forced cooling was required to keep the warranrty intact)

  3. Nigel 11

    An important question : SSD failure modes?

    What I'd really like to know about SSDs in general and SSD models in particular, is how they fail. Hard drives mostly (not always) give advance warning through the SMART parameters of when it is a good idea to replace the drive proactively. They also more often than not fail in such a way that ddrescue or suchlike are able to perform a complete or almost-complete copy of the contents.

    Apocryphally (or perhaps related to early products) SSDs turn into bricks "just like that". Perhaps only a few years of working with SSDs can fill this knowledge gap. Still, it would be nice if some review site could embark on a useful long-term endurance test. Write and rewrite flat-out while keeping an eye on SMART to see whether the drive gives warning of its own demise with days (or preferably weeks) to act on the warning.

    1. Paul Shirley

      Re: An important question : SSD failure modes?

      http://techreport.com/review/26523/the-ssd-endurance-experiment-casualties-on-the-way-to-a-petabyte

      Provided the controller doesn't die prematurely (used to be a problem) looks like you get sufficient warning from SMART to retire an SSD. I still won't be storing anything that cant be reconstructed or pulled from my system backup on one though.

      1. Ian Michael Gumby
        Facepalm

        Re: An important question : SSD failure modes?

        Can you say raid?

        I like RAID-10 so if I lose a drive, I can replace it and rebuild the raid group.

        And for redundancy... have a secondary server.

        1. tin 2

          Re: An important question : SSD failure modes?

          So I had 4 identical Samsung drives - not in a RAID at the time but could easily have been - fail one after the other in the space of a day. RAID is out of the window if that happens - and much more often than not RAID has identical drives? Perhaps it teaches me something about Samsung drives, but definitely back up, onto a variety of different manufacturer's stuff!

        2. Daniel B.

          Re: An important question : SSD failure modes?

          I like RAID-10 so if I lose a drive, I can replace it and rebuild the raid group.

          I prefer RAID-5 or RAID-6, as that doesn't cause me to lose 50% of my storage space to redundance. Also, rebuilding on SSDs won't have the same problem that HDD RAID has, namely that a second disk might fail while the first failure is rebuilding; mostly because SSD failures will usually be on writing instead of reading.

    2. Steven Jones

      Re: An important question : SSD failure modes?

      All the HDDs I had failed completely without warning. If people base their backup model on having warning of an impending failure then they are playing a dangerous game. Admittedly I've not had an HDD failure for a few years, but you have to be prepared for the worse.

      Anyway, with 10 year guarantees, perhaps we'll hear less about write endurance which, for the vast majority of uses outside of enterprise servers, is simply not a serious issue.

      1. Nigel 11

        Re: An important question : SSD failure modes?

        All the HDDs I had failed completely without warning

        I'm very surprised. Was this a spectrum of different drives from different vendors of different ages, or a (faulty!) batch of disks or systems all procured at once? And were you actively monitoring the drives' SMART parameters as they aged, or just waiting for a "replace drive" warning that never came?

        In my experience the reallocated block count (plus pending reallocation count) is the key parameter. It may be non-zero on a new drive, or rise from zero when the drive is slow-formatted (ie, every sector written to). It should then be on a plateau. If it is rising, beware. If it starts rising exponentially, swap that drive out asap.

        Yes, I have experienced hard drive controller failures (insta-brick) but this has been a rare event compared to more or less gradual degradation of the medium. And even when the controller has gone seriously wrong, ddrescue has not infrequently managed to rescue the data.

        You are right, backups are esential if the data is important. But where I work, there are many Terabytes of simulation output, kept for future reference. (A bit like the science-research equivalent of a PVR!) If the drive dies, so be it. The simulation can be re-run if necessary, at the cost of many core-hours of number crunching. The economics is such that it's better to use two multi-Tb drives to store 2x as many results for future reference, than to set the drives up as a mirrored pair.

        1. Anonymous Coward
          Anonymous Coward

          Re: An important question : SSD failure modes?

          "I'm very surprised."

          In the last three years I have had two users' 500GB hard drives fail within the warranty period. One was about a year old. The other lasted five hours - just long enough to have done all the Windows build, user data transfer, and updates ready to ship the PC. In both cases there was no warning - just dead on power up. It taught one user a hard lesson about using the back-up drive they had been given and had not used regularly when recording music sessions.

          One disk was a Seagate - the other Western Digital. I now only send out an upgraded PC with a well-used disk taken from my own PC.

          Out of the many hard drives that have failed over the last 10 years only about half gave any indication of a problem to allow pre-emptive action. Data recovery saved most of the files - and usually the disk was ok after using the manufacturer's repair utility. Only one disk ever gave a SMART warning of a deteriorating situation before errors were noticed.

          A Seagate USB external 1.5GB hard drive gave problems of repeated "click" retries on big transfers. This appeared to be well-known problem attributed to the Seagate USB implementation. Recently the disk was removed and put into an OEM USB enclosure where it performs without error.

        2. Steven Jones

          Re: An important question : SSD failure modes?

          I'm not talking abut enterprise arrays (the article was not about such things). Where I used to work we had storage measured in petabytes and databases in the 100s of GB. Those sort of devices are actively monitored and, in any case, use RAID protection. Pre-emptive swapouts happened, but complete failures happened too. Sometimes the pre-emptive swapouts were done because of unexpected failures over a number of drives and batch failures were present.

          No, I was referring to my own personal experience of consumer grade units. The monitoring of those is poor and I would hazard a guess that the vast majority of users never look at the smart stats and, in any case, wouldn't have much of a clue on what they mean. Also my experience dates back a long time when such stats didn't exist or were poorly monitored. The few clues you used to get were things like sticky bearings stopping the drive spinning up (a sharp rap with a screwdriver handle). The catastrophic failures tended to involve a lot of clicking or floods of I/O error messages followed by a locked up machine. Or a laptop that didn't work. (Or a PVR that stopped working).

      2. JEDIDIAH
        Mushroom

        Re: An important question : SSD failure modes?

        > All the HDDs I had failed completely without warning.

        I have only had one HDD fail without warning ever. I've had plenty announce their impending demise through SMART. Based on my own PERSONAL FIRSTHAND EXPERIENCE, depending on diagnostics from the drives themselves has not been a dangerous game at all. It's very useful even with notorious models from bad vendors.

        Self diagnostics on drives is highly useful. Poo-pooing drives that fail to provide this is unconvincing.

        Ignoring this feature is good enough reason to kick a particular SSD brand to the curb.

    3. Alan Brown Silver badge

      Re: An important question : SSD failure modes?

      > Apocryphally (or perhaps related to early products) SSDs turn into bricks "just like thatApocryphally (or perhaps related to early products) SSDs turn into bricks "just like that"

      I have several consumer-grade SSDs which like to lockup for no apparent reason (they come back when power cycled)

      > Still, it would be nice if some review site could embark on a useful long-term endurance test.

      Oddly enough, Anandtech and a couple of others have been doing exactly that for the last couple of years.

  4. Anonymous Coward
    Anonymous Coward

    AFAIK, SSDs are, and have been, more reliable than spinning HDDs, which are prone to arriving broken. 10 year warranty?

    1. Nigel 11

      AFAIK

      As fas as anyone knows. That's the problem. I'd far rather have media warrantied only three years but which can be expected to degrade gracefully thereafter, than something warrantied for ten years but with a greater risk of sudden and total failure.

      I'm old enough to remember some of the unfortunate ageing problems that have hit computing in the past. Purple Plague and Black Death of chips. The DEC disk drive where a beancounter substituted an adhesive marked "DO NOT SUBSTITUTE", and thereby doomed every disk coming off that production line to die from surface contamination within three or four years.

    2. JEDIDIAH
      Linux

      The myth of SSD reliability

      I have used a large number of HDDs. The devices are cheap and readily available in a way that SSD doesn't allow for yet. So claims about SSD reliability are mostly wishful thinking at this point.

      Plus, such claims also aren't bourne out by actual customer reviews.

      1. Bronek Kozicki

        Re: The myth of SSD reliability

        How much data can you write & erase on your HDD ? These guys stored quite a lot on a number of SSDs. And reported results for you.

  5. Ribblethrop

    ...in addition

    Glad to see you did not measure the speed the cache is read at and the quote that as the speed of the drive, much better than the last SSD review you did.

  6. James 100

    Smaller, faster?

    "the 240GB drive is a wee bit quicker at up to 520MB/s."

    Isn't this odd for SSDs? I thought the larger ones tended to be faster, because they're spreading the load over more chips.

    With decent backups, reliability isn't such an issue - I've had plenty of regular HDDs die on me over the years. The SSDs seem to be fine so far, but of course that's a much smaller and newer sample.

  7. AJames

    But can you rely on SanDisk's support

    I have a SanDisk SSD on the system I'm using, which I bought because of the great performance specs. Only to find a few months later that performance was deteriorating rapidly because of a firmware bug in the Sandforce controller. It turned out that every SSD manufacturer was aware of this bug and had agreed not to mention while awaiting a fix from Sandforce. When the bug was reported publicly, every SSD manufacturer other than SanDisk acknowledged the problem, apologized to customers, and released the fix as promptly as possible. SanDisk stonewalled customers, refusing to respond. Eventually they caved in and release a bug fix for their product, months after all the other manufacturers. I suspect that they would have quietly let it slide if the issue hadn't become public. Draw your own conclusions, but I'm avoiding SanDisk products in future.

  8. phil dude
    Boffin

    raid?

    Anyone out there using these for RAID? Since they are so much bigger, when will the death of RAID6 occur for SSDs?

    P.

    1. Anonymous Coward
      Anonymous Coward

      Re: raid?

      I am using PRO840 512GB in an 8 disk RAID-6 configuration - 1.8 used of 2.8TB

      The array is housed in an HP DASD cabinet connected via the standard p410i raid controller to a G7.

      This is of course an unsupported configuration as far as HP is concerned, but it was cheap and gave me an immediate capacity boost at moderate cost.

      I am not completely happy with this setup, as several of SSDs died for no apparent reason within a couple of months, which were RTMed without problem. Additionally, it is a lottery whether the RAID configuration will survive a power cycle of the G6, for reasons not at all clear, and in practice unknowable.

      Rebuilding the array after a single disk loss takes "forever", an attribute of the RAID controller I suspect.

      This array is for a moderately sized SQLServer database, so the minimum IO size is 8K and the maximum 1MB. These IO sizes are well above the target "optimum" for the SSD circuitry as I understand it, so some strange spikes do seem to occur.

      We have never had a database corruption though, so we are not totally unhappy. Our biggest issue is that when we actually need to trawl the data, we cannot get more than 800 odd MB/sec.

      I need more money :)

    2. Bronek Kozicki

      Re: raid?

      I'm running 8 Crucial C300 (yes, old model) 256GB each on LSI MegaRAID in RAID0 hardware configuration, total 2TB. It's nice and no problems in years since I set it up, but in fairness I do not write this much data to it. Of course, I back it up almost constantly to an HDD and replicate the backups to external HDD, although so far the only use of backups was deleted files or misconfigurations.

  9. psychonaut

    ssd's

    ive used many diferent types. im a system builder (among other things).

    havent used anything other than samsung 840 pro's for a year or so now.

    the ocz, sandisk and corsair drives i tried had a huge failure rate, and yes its the "bbbllrt...dead" failure type. the ocz's were the worst, about 50 % died within 3 months. luckily i only used about 6 of these before it became apparent. the samsungs i think ive only had 1 failure so far and ive probably built close to a 100 pc's with them in the last year or so.

    just installed the first 850 pro in a new build...blimey its fast. the 850's are also now cheaper than the 840 pro's. rapid mode also works on the 850's, it really is awesome.

    i hope the sandisks are more relaible now than they were a year or so ago, but i wont be going back to find out.

  10. Ian 55

    On disk encryption

    This may be me being dim, but doesn't this just mean is that if the controller dies, so does your data?

    Someone with access to the working drive can still get at the contents, but if it goes wrong then - as with the external drives with this 'feature' - you're stuffed.

  11. a4552597

    It's a big positive having no hardware encryption

    The writer implies that having no TGC Opal 2.0 is a bad thing but I disagree. I'm always going to set up my OS to encrypt in software at the block level. Why would I want the SSD to do this as well - it's more complexity, less transparency (how can you trust the manufacturer's implementation) and if the SSD corrupts it's master key your toast. With software block level encryption any corruption any where on the SSD and I just lose that one block. I really despise this TGC Opal 2.0 stuff - you can't trust it, there's more failure modes and yet it's imposed on us even when we don't want it.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like