Samsung SSD 840 Test Results, and Questions.

Discussion in 'SSDs & Data Storage' started by interested, Nov 29, 2012.

  1. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    Hello everyone.
    I have been building a PC for a client, which has a Samsung SSD 840 with 6B0Q firmware, installed as the OS drive.
    And i have some performance results that i would like to share, and ask if others are seeing the same performance on their Samsung SSD 840 drive.

    First off here is the windows 7 pro 64bit SP1 system (fresh clean install) that this drive is connected too.
    This SSD is connected to an Asus Sabertooth X79, socket 2011 motherboard.
    Plugged into the Intel 6Gbit ports, and setup in bios for AHCI operation, and write cache enabled in windows device manager.
    Newest drivers on Asus download page were installed on this board, such as.
    Intel chipset driver version 9.2.3.1025
    Intel Rapid Storage Technology enterprise Driver Software V3.5.0.1092.
    Intel Management Engine Interface V8.1.10.1275

    The samsung SSD magician software ver 3.2 was also installed.

    As part of this build i have so far run two programs to verify the performance specifications of the samsung SSD 840 which are advertised as follows:
    Advertised Sequential Read Speed:Up to 530MB/s
    Advertised Sequential Write Speed:Up to 250MB/s
    Advertised Random Read Speed:Up to 96K IOPS
    Advertised Random Write Speed:Up to 62K IOPS

    The first thing I ran was the Samsung SSD magician v3.2 software that came with the drive. I ran the OS optimization section, which basically disables disk defragmenter, drive indexing and checks that windows is setup for SSD use, then i ran the benchmark.

    The results from this benchmark are as follows:
    Seq Read: 532MB/s
    Seq Write: 102MB/s
    Random Read:92K IOPS
    Random Write: 21K IOPS
    [​IMG]

    The read numbers match advertised, however the Write numberes look awful.

    I was wondering if anyone out there has run this Samsung Magician benchmark on their Samsung SSD 840 drive, and if they see the same results?

    I went ahead and also ran ATTO disk benchmark which gave me the following results which are closer to what is advertised
    Max Read = 564MB/s
    Max Write = 249MB/s


    [​IMG]

    I wonder why the samsung magician app reports different values for seq write compared to ATTO?

    I wonder if anyone else out there sees the same kind of results?
    Also does anyone have any tips on anything I should do to try and improve the performance of this drive on this system?

    I have noticed a few small pauses when doing things winthin windows, things like moving files, installing apps, or even loading into windows, which makes me wonder if maybe i should be using a different driver, or if there is some tweak i should perform, or if the drive itself has issues.
     
    Last edited: Dec 3, 2012
  2. Brahmzy

    Brahmzy [H]ardness Supreme

    Messages:
    4,955
    Joined:
    Sep 9, 2004
    I'd stay away from that Magician benchmark thing. Post up some AS-SSD and Anvil runs.
    The small pauses are not right. I've got a single 128 Pro running on a P67 box of mine and it's balls-out fast. How full is the drive? Did you partition out the whole drive or leave some OP room?
    I'm also not as familiar with the correct X79 RST drivers as I am with the 77/67 drivers.
     
  3. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    I will perform the AS-SSD benchmark later on tonight, i am not familiar with Anvil but i will look for it, and run it.

    I am still interested in seeing other peoples results when they run this samsung magician software, just to confirm that the magician software is indeed not accurate or problematic, so if you get a chance can you run it on your system and see if it returns advertised speeds?

    The SSD is brand new, and it has a clean fresh install of windows 7 pro 64bit on it, with one standard NTFS partition ( plus the small extra partion that windows 7 usually creates).
    The drive is basically empty other than the windows 7 install that resides on it. ( i can include actuall numbers later on tonight).

    I just noticed that there is a slightly newer RST driver available at intel
    I have Version 3.5.0.1092 installed, and there is one that has a slightly higher number of 3.5.0.1101
    I might try installing the newer one to see if it improves anything.
     
    Last edited: Nov 29, 2012
  4. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    I was able to run AS SSD, and Anvil, here are my results.

    [​IMG]

    [​IMG]

    The AS-SSD sequential numbers were similar as advertised, however the Anvil results were similar to the low samsung magician numbers.

    At this point i now had two benchmarks showing poor performance on this drive (samsung magician and Anvil).
    So i figured something was either not properly setup, or there are issues with this drive.
    One thing that i tried next was to go over my checklist of SSD configuration items.
    Thats when i noticed that while i had enabled write caching on the SSD, i did not turn off the "windows write-cache buffer flushing on the device"

    [​IMG]

    So i disabled the windows write-cache buffer flushing on the device, rebooted, the pc and re-ran my benchmarks, and here are the results.

    [​IMG]
    [​IMG]
    [​IMG]

    So with the windows write-cache buffer flushing disabled, i am now getting much higher write numbers than i did before, which are closer to advertised values. (IOPS are still low but at least closer than before)
    Even the Samsung magician benchmark is now reporting better results.
    The read numbers went down a little bit with this adjustment, however the large bump in write values makes up for the small decrease in read.

    Now I know this is not an enterprise grade SSD, so its likely not to have a capacitor required to prevent issues during power failures, therefore I don't know if I should leave the SSD drive setup this way, for the aditional write performance?

    Also i am not aware if SSD manufacturers expect people to run windows with the write-cache buffer flushing disabled on a consumer grade drive in order to acheive their advertised speeds?

    If so why did the Samsung Magician software not have disabling the write-cache buffer flushing as part of its OS optimization steps? Other common SSD setup steps were included in the optimize section of the magician software (such as turning off disk defragmenter, ready boost, and super fetch etc..).

    Is everyone else here also disabling their windows write-cache buffer on their non enterprise SSD setups, in order to get advertised speeds, or do you get your advertised speeds with write-cache buffer flushing enabled?
     
  5. Brahmzy

    Brahmzy [H]ardness Supreme

    Messages:
    4,955
    Joined:
    Sep 9, 2004
    All depends on your risk tolerance. I have a pretty healthy backup routine for all of my stuff. If it was for a customer/friend/family? Not a chance.
    Also keep in mind the manu. gives its specs based off of a 100% clean spare drive, not one with an operating system running on it.
     
  6. JoeComp

    JoeComp [H]ard|Gawd

    Messages:
    1,036
    Joined:
    Jan 23, 2012
    The other thing to keep in mind is that you are running on an X79 system. For some reason, the performance of SSDs on those systems is strange, especially with regard to write cache settings.

    On a Z77 or P67 system, most people (including myself) can get excellent SSD benchmarks without disabling buffer flushing.
     
  7. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    This is a very interesting comment.
    I was not aware that an X79 system would have such strange issues.
    I just googled this subject and apparently others are seeing this issue too, where they have to disable write-cache buffer flushing to get advertised speeds.

    It is likely that this is probably the issue i have as well, since i would find it very unlikely that Samsung and other consumer grade SSD manufacturers would expect people to disable write-cache buffer flushing in order to get advertised speeds (either used as OS drive or spare)

    In one forum I read that using the standard RST (non-enterprise) version might fix this issue, so thats somthing I will look into.
    Unfortunately the last post in that forum was posted months ago, and the version number listed was quite different than the latest version of RSTe that i downloaded from intel's website. So it looks like i have to do some trial and error testing with various versions of RST or RST(e) to see if that helps.

    You would think that intel would have already fixed this issue by now on the newest RST(e) since x79 has been out for some time now.
     
    Last edited: Nov 30, 2012
  8. bAMtan2

    bAMtan2 [H]ard|Gawd

    Messages:
    1,479
    Joined:
    Sep 16, 2008
    This definitely sounds like a write cache issue. Flip the switch or not. There is no middle ground. Either you don't mind the risks of data loss with power loss, or you do. People who don't will tend to buy uninterruptible power supplies with automatic voltage regulation to prevent problems from that area
     
  9. ToddW2

    ToddW2 2[H]4U

    Messages:
    4,019
    Joined:
    Nov 8, 2004
    Which is a rather cheap investment, and helps in a lot more scenarios than data loss too :D
    $100-$200 for a UPS (ON SALE), $100 or so for a voltage regulator on amazon.

    I run all my electronics through the VR -> UPS -> Electronic Device. (Routers, Stereos, Computers).

    The rest of my house I use VR on important electronics. (Look at the specs of the regulators some will correct 5% others up to 10%)

    -Todd
     
  10. extide

    extide 2[H]4U

    Messages:
    3,297
    Joined:
    Dec 19, 2008
    Looks like your alignment is off to me. I highly doubt it is the write cache setting.
     
  11. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    Curious. Can you explain why you think this?
    The anvil benchmark pics show the alignment as being "OK" with cluster size of 4096B

    Also i dont know how it would be improperly aligned given that it was a brand new blank SSD, and windows 7 pro 64bit was then installed on the blank drive from scratch.
    While I did not manually define my partition during the clean install (as done in this video), I just let windows use its default partition, sectors and file system which i figured would already create a 4K NTFS partition.

    I did not transfer windows (migrate/mirror) onto this SSD from a pre-existing install.

    Im open to suggestions if you do think its alignment. is there any other tool i can use, or steps i can follow to confirm your suspicion?

    I think that the benchmarks clearly show a difference where turning off write-cache buffer flushing, increased performance to values similar to that advertised by samsung.
    So maybe it is an RST related issues?
     
    Last edited: Nov 30, 2012
  12. extide

    extide 2[H]4U

    Messages:
    3,297
    Joined:
    Dec 19, 2008
    Well, maybe it's not alignment then. Those read/write speed patterns just looked like bad alignment to me.
     
  13. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    I did some more testing on the weekend whereby i tried different RST and RSTe drivers.

    I tried the last 2 releases of RSTe taken from the intel website, (3.2.0.1126 & 3.5.0.1092/3.5.0.1101) and they both had the same issue whereby I had to disable write-cache buffer flushing in order to get decent write speeds.

    Then i tried the regular RST version 11.6.0.1030 (even though X79, with the C600 storage is supposed to use RSTe).
    Here are my Anvil and Samsung results, with the write-cache buffer flush enabled (unchecked).
    [​IMG]
    [​IMG]

    So it appears as if it was the RSTe driver that was holding back on the performance of this SSD drive.
    Use of the standard non-enterprise RST driver finally has this system working near advertised performance specifications, with write-cache buffer flushing on (unchecked).

    Here is another screen shot using the samsung ssd magician tool comparision showing RSTe vs RST driver both with write-cache buffer flush enabled (unchecked).
    [​IMG]
    Blue = RSTe, Red = RST

    So now i can have this clients machine operating with the write cache buffer on (for safety), and get the 840's SSD advertised speeds.

    I would like to thank JoeComp for opening my eyes to the X79 SSD issue, which i read up on a few other forums, and which eventually lead me to my solution of using the RST driver instead of RSTe.

    It is too bad that Asus never bothered to test their Sabertooth X79 board with a SSD drive, because if they did they would have noticed the same thing, and then could have at least provided the RST driver on their download page, (which still only contains RSTe) with a readme telling people to use it instead of the RSTe driver.
    This would have saved me all this testing.

    TL;DR The lesson here is if you have a X79 motherboard, use the intel RST driver not the RSTe driver to get advertised SSD speeds with write-cache buffer flush on, or buy an enterprise level SSD so you can safely run with the write-cache buffer flush off.
     
    Last edited: Dec 3, 2012
  14. JoeComp

    JoeComp [H]ard|Gawd

    Messages:
    1,036
    Joined:
    Jan 23, 2012
    Your 4KQD1 reads are low. The 840 Pro is particularly good at those, and should get in the 35-40MB/s range, at least on P67 or Z77 systems with ASU and that version of IAStor.
     
  15. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    This is a standard 840, not the Pro version.
     
  16. JoeComp

    JoeComp [H]ard|Gawd

    Messages:
    1,036
    Joined:
    Jan 23, 2012
    Aha! I missed that. Thanks for pointing it out.
     
  17. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    It is however still a bit lower than what i see on reviews of the 840 drive (like on this review for example)
    That review shows 31.1MB/s and 110MB/s for 4K and 4KQD4 reads. where i get 24.3MB/s and 89.02MB/s
    They also obtain 119MB/s and 235MB/s for 4K and 4KQD4 writes, where i get 58MB/s and 238MB/s.

    However the review sites are using a clean 100% free drive on a different chipset with different RST, where mine is 77% free and is on x79 with a different RST.

    So yes overall i am still not getting the absolute best I can out of this SSD drive, however it is now close enough.
    At least now both read and write performance values are closer to advertised.

    One thing i still have to check is if TRIM is still enabled now that im using RST instead of RSTe. Hopefully it still is.
     
    Last edited: Dec 3, 2012
  18. dismas

    dismas n00b

    Messages:
    3
    Joined:
    Sep 19, 2012
    The read numbers for my Samsung 840 are really low compared to yours. What could be wrong? I'm using Windows 8 Pro on a MSI H61M-P21 (B3) motherboard.

    Samsung SSD Magician benchmark results:
    [​IMG]

    AS SSD benchmark results:
    [​IMG]

    Anvil benchmark results:
    [​IMG]
     
  19. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    Your anvil results state that your using MSAHCI? driver.
    I don't know what driver that is?
    Are you plugged into the intel chipset SATA ports or some other 3rd party chipset?
    If so try plugging your drive into the intel 6Gbit ports and then get the 11.6 RST driver from intel.
     
  20. Brahmzy

    Brahmzy [H]ardness Supreme

    Messages:
    4,955
    Joined:
    Sep 9, 2004
    He posted on anand that he was accidentally using a SATA2 port. He's all good now.
     
  21. dismas

    dismas n00b

    Messages:
    3
    Joined:
    Sep 19, 2012
    Hmm I'll have to double-check when I get home today.

    Are you talking about me? Pretty sure that wasn't me :eek:
     
  22. Brahmzy

    Brahmzy [H]ardness Supreme

    Messages:
    4,955
    Joined:
    Sep 9, 2004
    LOL - there was another guy who had the exact same problem and looked to have the exact same screenshots and the same day of posting, LOL.
     
  23. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    BTW in case anyone was still wondering, TRIM is still enabled on the non-enterprise RST.
    I checked it but forgot to report back.
     
  24. interested

    interested Gawd

    Messages:
    795
    Joined:
    Feb 3, 2006
    New firmware released by samsung for the 840 (DXT07B0Q).
    This is what it says on the samsung webpage:LINK
    "This update addresses the following issues :
    1. Improved "dirty drive" write performance."

    Has anyone tried it yet?
     
  25. Cov

    Cov Gawd

    Messages:
    917
    Joined:
    Dec 3, 2007
    Yes, I benchmarked, updated the firmware and bechmarked again.
    Results:

    Samsung 840 Pro 256GB
    Firmware: DXM03B0Q ◄ LEFT and DXM04B0Q RIGHT ►


    [​IMG] . [​IMG]

    [​IMG][​IMG]

    [​IMG][​IMG]

    Specs

    CPU: i5 3570K
    Mainboard: Intel DH77DF
    RAM: 16GB G.Skill
    OS: Windows 7 x64 Ultimate
     
  26. Cov

    Cov Gawd

    Messages:
    917
    Joined:
    Dec 3, 2007
    Here is the benchmark result of Anvil's Storage Utilities after the Firmware update:

    [​IMG]
     
  27. JoeComp

    JoeComp [H]ard|Gawd

    Messages:
    1,036
    Joined:
    Jan 23, 2012
    Note that those are 840 Pro benchmarks, not 840 like the OP was talking about.