Best SSD raid benchmark utility?

Discussion in 'Mac Pro' started by sbab, Aug 3, 2015.

  1. sbab macrumors member

    Joined:
    Oct 25, 2012
    #1
    MACPRO 5.1 2010
    What is the most accurate benchmark utility that I can use to test a raid 0 setup of a sonnet tempo pro with two samsung 840 pro 512 gb ssds?
    I have used blackmagic, quickbench and diglloyd disktester with varying results.
    Blackmagic has given lower results than the other two especially the write speeds.
    I don't test for test sake but my raid isn't performing so well and I'd like find out just how good or bad it is.
    Many thanks
     
  2. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
  3. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #3
    trim is enabled with disk sensei plus a manual trim also carried out
     
  4. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
  5. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #5
    This is what I have from a single 840 Evo 1T SSD on the Sonnet Tempo card (not the Pro card).

    DiskSpeedTest SSD.jpg

    Very close to the speed declared by Sonnet.

    Screen Shot 2015-08-04 at 17.08.35.jpg

    So, it seems that the Blaskmagic is quite an accurate tool to measure the Tempo + 840 series SSD setup.

    For the Pro card in RAID 0, you may get up to the following figure.

    Screen Shot 2015-08-04 at 17.10.52.jpg

    And what's your result from Blackmagic now?
     
  6. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #6
    These results were obtained with two samsung 840 pros 512GB setup as raid 0 on sonnet tempo pro.
    Raid 0 is 18 months old and trim enabler had been activated. Raid was benchmarked when 10.10.4 came out, results were <100 write / about 200 read. I bought disk sensei to try and regain performance. Raid 0 is only half full.
    The raid 0 was manually trimmed with disk sensei and also the "fsck -fy" command and this is the result.
    samsung raid sonnet.png
     
  7. pastrychef macrumors 601

    pastrychef

    Joined:
    Sep 15, 2006
    Location:
    New York City, NY
    #7
    Blackmagic Disk Speed Test has always seemed to work well to me...

    Screen Shot 2015-08-04 at 6.15.41 AM.png
     
  8. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #8
    The result looks very wrong to me. The 500+ reading speed confirms that's a RAID 0 config on the Tempo card, otherwise it's limited to 500. However, 555 is quite low for a 840 Pro RAID 0 setup. And the write speed is even not close to my single 840 Evo, that's very wrong.

    Did you use the 1G test size? Will any other size made the difference?

    Also, if this RAID 0 is a new setup, may be you can just separate the SSD and benchmark them one by one to see if any one of them cause the problem. (Of course, it may not be a good idea if the data is already there)
     
  9. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #9
    For the blackmagic tests I used 5GB, I think thats what people recommend.
    The raid 0 is not new it was started in jan 14. Other than the initial test when new it hadn't been tested until the time support for 3rd party trim was introduced.
    In the last day I have tested the ssd with some different bench testers the results all reflected the blackmagic score, low overall results and poor write performance.
    The other utilities I used were AJA 180W/595R - disk sensei 153W/483R and terminal 167W/432R
    The problem are the ssd's, so thanks for the help in getting me one step closer.
     
  10. SoyCapitanSoyCapitan macrumors 68040

    SoyCapitanSoyCapitan

    Joined:
    Jul 4, 2015
    #10
    AJA squeezes out a bit more Performance than Black Magic.
     
  11. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #11
    It's a bit hard to find out exactly what's causing this poor performance.

    However, for this kid of performance, IMO, there is no point to keep this RAID 0 config.

    If I were you, and I setup this RAID 0 for performance, but not treat it as a bigger SSD. I will clone all data to the backup disk (If the data is important, use at least 2 different method to backup the data to 2 physically separated drives). Break the RAID 0, test the SSD one by one, and then decide my next step.
     
  12. pastrychef macrumors 601

    pastrychef

    Joined:
    Sep 15, 2006
    Location:
    New York City, NY
    #12
    These benchmarking utilities are used to measure performance, not to squeeze out more performance.
     
  13. SoyCapitanSoyCapitan, Aug 6, 2015
    Last edited: Aug 6, 2015

    SoyCapitanSoyCapitan macrumors 68040

    SoyCapitanSoyCapitan

    Joined:
    Jul 4, 2015
    #13
    But...AJA does squeeze out bit more performance ;)

    CrystalDiskMark on Windows even more.
     
  14. netkas macrumors 65816

    Joined:
    Oct 2, 2007
    #14
    Nah, this blackmagic test shows maximum seq read/write speeds, the thing that rarely happen in real life.

    good old xbench shows numbers closer to real usage patterns
    For example evo 850 over sata2
     

    Attached Files:

  15. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #15
    I agree with that. However, it means Blackmagic is good for checking if the setup can achieve what the manufacture said. Isn't it? Because they always use that max sequential R/W to indicate performance.

    I think it will be quite rare that Blackmagic shows bad number but xbench shows good number.

    Anyway, here is my result (single 840 Evo on Tempo SSD) for OP to compare his result.

    Screen Shot 2015-08-06 at 19.03.10.jpg
     
  16. SoyCapitanSoyCapitan macrumors 68040

    SoyCapitanSoyCapitan

    Joined:
    Jul 4, 2015
    #16
    I have posted benchmarks screenshots in the NGFF of vanilla SM951.

    Blackmagic was hitting little under 1400. AJA was hitting around 1430. CrystalDiskMark was little over 1500 - the max that pcie 2 can manage with the ssd.
     
  17. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #17
    The benchmark utilities indicate a ssd problem, bugger!
    I used disk sensei to turn trim on and have carried out a manual trim. I have also used the terminal command "fsck -fy". They improved the performance up to the blackmagic test I displayed.
    Is there anything I can do to restore the performance?
    As mentioned above I could break the raid then test each ssd individually but before going to that extreme is there anything else quicker easier?
    If it gets to breaking the raid and the test results are low, what do you do next; erase, secure erase and/or add a small extra partition to the ssd for trim use? Can someone please help me out here.
     
  18. hfg macrumors 68040

    hfg

    Joined:
    Dec 1, 2006
    Location:
    Cedar Rapids, IA. USA
    #18
    This was my boot drive prior to adding a Apple PCI SSD and consisted of a pair of 256GB Samsung 840 Pro SSDs mounted on a Sonnet Tempo Pro PCI card in a Mac Pro 5,1:

    840ProRaid0SonnetPCI.png
     
  19. Cindori macrumors 68040

    Cindori

    Joined:
    Jan 17, 2008
    Location:
    Sweden
    #19
    Have you tried benchmarking in Disk Sensei?
    Curious to see how it compares to Blackmagic.
     
  20. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #20
    Yes I have and I'll test again.
    Here are Disk Sensei and Blackmagic tests, same time one after the other:
    DiskSpeedTest.png sensei test.png
    PS: When I select the "screenshot" button in Disk Sensei it "quits unexpectedly" I'll try to send you the report
     
  21. Loa macrumors 65816

    Loa

    Joined:
    May 5, 2003
    Location:
    Québec
    #21
    Hello,

    Don't know if it's been said before, but the best benchmark is the real world. You usually need to transfer huge files? Create a folder with 3-4 of your usual files and time the transfer. You usually need to transfer a lot of small files? Copy the biggest folders of such file you have, and time it.

    Why mess around with unreliable benchmark utilities when you can actually time what you actually need to do???

    Loa
     
  22. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #22
    Because it's hard to compare the real world result with the others / manufacture's spec.
     
  23. sbab thread starter macrumors member

    Joined:
    Oct 25, 2012
    #23
    Problem solved...I used diglloyds disktester.
    I broke the raid and tested each ssd, I got the same poor results
    I erased and used the "ssd recondition" selection on disktester.
    The ssd's then tested individually ok so I put the raid back together and cloned the raid 0 from the backup.
    Here is the result... but I don't know for how long, fingers crossed.
    raid 32 master loaded.png
     
  24. Cindori macrumors 68040

    Cindori

    Joined:
    Jan 17, 2008
    Location:
    Sweden
    #24
    If you just measure the time it takes to copy a file, you don't measure the speed of drive. You measure the speed of the filesystem. And the next copy may be 500% faster because of caches. Or 20% slower because the file is another data type. Benchmarks go through great lengths to ignore those things and actually measure the speed of the drive.
     
  25. Loa macrumors 65816

    Loa

    Joined:
    May 5, 2003
    Location:
    Québec
    #25
    Well, if you don't have access to the drives yourself, it's hard to use benchmarks as systems (the machine itself, the processes that affect read/write speeds, drive fragmentation on source or destination, etc...) will differ significantly anyway.

    benchmarks will give you ballparks. If that's not enough, then you're out of luck or the drives are already very similar.
     

Share This Page