OWC Mercury Accelsior 480GB PCIe SSD Review – First Upgradeable PCIe SSD Hits The Streets


Crystal Disk Benchmark is used to measure read and write performance through sampling of raw (0/1 Fill/compressible) or random data which is, for the most part, incompressible. Many new ‘SandForce Driven’ SSD owners who can’t wait to test the performance of their SSD often grab this program and run a quick test, not realizing that they are testing with incompressible data rather than compressible data used in testing by manufacturers.  We have provided compressible (oFill) results on the left with incompressible (random data) results on the right.

blankAlthough there is a visible difference between the sequential and 512k writes when comparing both results, it is not nearly as high as expected and, in fact, we can attribute the relatively high transfer speeds of above 500MB/s, while testing in highly compressible data on the right, to the hard work of the Marvel controller.  In the same respect, the low 4k randoms write performance is a bit lower than expected which, once again, would point to the RAID controller and configuration.  Testing the single blade drive alone would probably display a low 4k result of around 85-90MB/s which has become the norm for similar ‘SandForce Driven’ SF-2281 SSDs.


Up until recently, AS SSD was the only benchmark created specifically for SSD testing and it uses incompressible data.  AS SSD, for the most part, gives us the ‘worst case scenario’ in SSD transfer speeds while using ‘SandForce Driven’ SSDs as they use compression in storage as discussed earlier.  Many enthusiasts like to benchmark with AS SSD for their needs.

blankblankOnce again, the definite strength we never expected to see lies in the sequential write performance on the left and the achilles heal just might be this 4k score once we test in PCMark Vantage.  This is the first we have seen the IOPS results and, considering that this is a consumer and small business SSD, 73181 write IOPS is nothing to complain about.  Lets see how this translates when moving data:

blankThe AS SSD Copy Bench provides us with the high transfer speed reached and time to move a file from one part of the SSD to another, the three data samples being that of an iso., Program and Game.  The ISO result is the fastest transfer we have seen to date while the transfer of the Program was a bit lower than expected.


Over the last little while, we have been assisting with beta testing new benchmark software called Anvil Storage Utilities which is an absolutely amazing SSD benchmarking utility.  Not only does it have a preset SSD benchmark, but also, it has included such things as endurance testing and threaded I/O read, write and mixed tests, all of which are very simple to understand and utilize in our benchmark testing.

blankWe found the results of Anvil Storage Utilities similar to what we have been getting thus far except the high sequential read speed is very low when considering both the listed specifications and previous results we have seen in ATTO, Crystal DiskMark and AS SSD.  It is not abnormal to find the odd compatibility issue when testing newly released technology.


  1. blank

    Is it bootable?

  2. blank

    can i use trim comand? does it have a garbage collector? after intensive use of the drive will it perform slower?

    • blank

      No TRIM would not work with this RAID 0 configuration, however, Garbage Collection would do its job with the SandForce drives. Unfortunately, the product doesn’t remain with us for a long period so any degradation cannot be measured through continuance use and analysis.

    • blank

      no, yes, yes

  3. blank

    Does the drive maintain the same performance figures in its smaller configuration (120GB)?

  4. blank

    It seems wrong to call this card a PCIe SSD card as such since it is really a Raid/HBA card which happens to accept mPCIe SSD’s on the board itself. And just why wasn’t this card manufactured with the ability to install 4 mPCIe SSD’s from the get go?

    • blank

      The Accelsior uses the Marvell 88SE9230 which only supports PCIe 2.0 x2 which only allows up 1000 MB/s. You would need to use a chip with 4 lanes (such as the Marvell 88SE9445) to get the full performance out of 4 SSD’s. Maybe that chip doesn’t have the boot ability that OWC advertises.

  5. blank

    Would you happen to know/offer any card or adaptor to connect an SSD like this (https://www.toshiba.com/taec/news/press_releases/2010/memy_10_604.jsp) and retrieve the data? It is the SSD from an Asus UX31 and I think the connection is similar to the ones of the SSDs in the newer versions of the Macbook Air.

  6. blank

    Please run CrystalDewWorld’s DiskMark with a larger test size. At 1000MB, the filesystem cache can skew results upwards. For example, running DiskMark w/1000MB on my system shows read/write operations around 480MB/s. Cranking up the test size to 4000MB returns a more realistic 290MB/s — appropriate for a SATA II SSD drive.

  7. blank

    I’m a bit puzzled as to exactly a PCIe SSD card does for a computer. I have an older Intel MAC Pro tower used mostly for audio recording and am trying to extract a bit more performance from the box during in-the-box multi-track audio recording. Does this device essentially only increase the read/write performance between the computer and external drives or are there other benefits? TIA

  8. blank

    spotted something that needs correction in your review- SandForce SF-2281 is a SATA controller not PCIe, and the Marvell RAID controller is also a SATA part. Not sure where you got the idea that the SSDs in this review were PCIe, but they are not.

    • blank

      The review is correct and it is not stated that the SF-2281 is a PCIE controller. This package is a PCIE product, in that, it is connect d via this route. Thanks anyway.

      • blank

        Thanks for your quick reply- however your review states
        ‘There are three detachable components which are the PCIe itself and two upgradable ‘blade’ mPCIe SSDs that can be switched off for higher capacity.’
        You have clearly referred to the SSDs as mPCIe which is not correct. Thanks anyway.

      • blank

        I am curious what validation you might have to prove this. Considering both the RAID controller and SSD controller are both PCIE compliant, your opinion should be validated in some way. Simply because the connector of the blade SSDs is custom, does not mean the term still cannot be used for best explanation to the reader, similar to the way one compares an Apple SSD to that of the M2.

      • blank

        The use of mPCIe was used as the best explanation at the time for the custom connector. While you are correct that the controller is SATA on each blade SSD, this custom solution is a PCIE design. Thanks again.

Leave a Reply

Your email address will not be published. Required fields are marked *