*Official* NBR Desktop Overclocker's Lounge [laptop owners welcome, too]

Discussion in 'Desktop Hardware' started by Mr. Fox, Nov 5, 2017.

  1. iunlock

    iunlock 7980XE @ 5.4GHz

    Reputations:
    1,879
    Messages:
    4,355
    Likes Received:
    6,097
    Trophy Points:
    581
    We're all curious to know, because there definitely is a difference. :)

    btw can you email me some of your download and upload speeds? lol
     
    jaybee83 likes this.
  2. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,126
    Messages:
    6,652
    Likes Received:
    5,798
    Trophy Points:
    681
    Yeah, I'm not overly influenced by tech sites, I do read reviews though & see if what they test stacks up to be good rigorous testing, most of the time the testing is good, but I don't believe in the inflated TBW endurance figures of the Phison E12 Toshiba TLC drives - which I mentioned before I think. Yes, my testing on my Sabrent Rocket 1TB drive is 'real world' testing, in as much it was a copy and paste of a dummy 170GB file from one folder to another folder on the same drive - you have to make sure it's "copy & paste" rather than "move", because "move" will just reassign the file attributes to the new folder rather than actually writing the data to the drive again - choosing "copy & paste" will instead actually write a copy of that file into the NAND, so that's a proper test of the write speed of a drive.

    I mentioned in my earlier post how I created a dummy file. I used fsutil file createnew Testfile170GB.exe 182536110080 command in a command prompt to create a dummy test file (you'll need some kind of experience with DOS or the command line to do that, minimal). Using the command line you have to navigate to the directory where you want to create the file, and then you just type in that command I gave you - as it will create the file in the directory that you are currently working in within command line. I then just used windows explorer to copy & paste the newly created 170GB file to another folder on the same drive - you know the simple right click the file & choose copy, and then right click the folder you want to paste it to & choose paste (ha!). Then you get to see the transfer speed as it copies - just like you can see in my attached pics in my earlier post.

    This Sabrent drive definitely outpaces all the reviews I've seen of any other TLC drive in this 170GB file transfer test - like I mentioned the Samsung 970 Evo and Evo Plus crap out to 'slow' write speeds at well under the 100GB write point - whereas my Sabrent drive was just writing at a constant 2.3GB/s the entire time for that 170GB file - I'm impressed, I think that's even faster than the 970 Pro, but I'm gonna Google that now.
    EDIT: actually it's about the same speed as a 1TB 970 Pro. (https://www.kitguru.net/components/ssd-drives/simon-crisp/samsung-970-pro-1tb-ssd-review/16/). A 100GB data file was written to 970 Pro at 2561MB/s which is 2.5GB/s, so that's only a little faster than the 2.31GB/s I saw with my cheap Sabrent Rocket drive.

    Hi Jaybee, I answered that in this post I'm typing right now.
     
    Last edited: Jun 29, 2019
    tilleroftheearth and jaybee83 like this.
  3. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,100
    Messages:
    11,564
    Likes Received:
    9,127
    Trophy Points:
    931
    hmmm, this is actually an interesting question. i never thought about how the SLC cache might come into play / how it behaves when u do file copy operations onto the same drive. could u do a comparison with a non-artificially generated file? say, a large movie file or smth like that?
     
  4. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,126
    Messages:
    6,652
    Likes Received:
    5,798
    Trophy Points:
    681
    Unfortunately I don't have any large 'real' files, which is why I created a dummy file, and I'm not gonna download a 170GB file for testing purposes, ha! Point is, the Samsung Evo drives crap out at well under the 100GB write mark even with a dummy file, whereas my Sabrent drive stayed at the same write speed for the entire 170GB file transfer. That's got to be a win for the Sabrent, I like it how the cheap underdog is taking names, ha!
     
    tilleroftheearth likes this.
  5. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,100
    Messages:
    11,564
    Likes Received:
    9,127
    Trophy Points:
    931
    thats the thing though, usually copy testing during reviews is either done with a second drive or against a RAM drive to avoid any bottlenecks :) that's why this is hard to compare against other data...
     
    Robbo99999 likes this.
  6. User32

    User32 Notebook Deity

    Reputations:
    43
    Messages:
    719
    Likes Received:
    194
    Trophy Points:
    56
    Thoughts on that """"""power outage""""" in the Toshiba SSD plant?
     
  7. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,126
    Messages:
    6,652
    Likes Received:
    5,798
    Trophy Points:
    681
    Copying & pasting a dummy file from one point to another on the same drive did not cause any bottlenecks from a Read point of view, as you can see from my HWInfo screenshots in my previous posts (not this one below, that's different) - the Read Activity was under 100% mostly and very low generally, while the Write Activity was fairly constant at close to 100%. I already posted a link to dummy file transfer testing within the same drive with Samsung Evos, and they crapped out at below 100GB. Here's this link I'm talking about, and one I posted earlier: https://www.tweakguides.com/Hardcon19_7.html

    I agree though that ideally you want to be transferring from a seperate drive that reads faster than your drive you are testing can write. In fact, I did some testing just now (see first pic below) with a 170GB dummy text file that is created that contains random information, which is different to the fsutil command line dummy file that just contains zeros. What I saw was that the Read Activity was at 100% while the Write Activity was fluctuating and at about 50% on average. I therefore concluded that this is not showing the true write speed because the drive is too busy trying to read the data (see following screenshot). I created this different dummy file using this program (https://www.mynikko.com/dummy/), and it's something about the random data within the file that means that the Read activity of the drive is the bottleneck in this case - needless to say the transfer speeds were quite low/embarrassing!
    170GB random text file transfer.jpg
    To properly test transfer of this random data containing 170GB text file it would have to be on a seperate drive that can read it faster than the receiving drive could possibly write - it seems like this randomly generated text file is a bugger to read fast, as you can see read activity is at 100%).

    I'm not sure what to make of my previous dummy file transfer speeds in my earlier posts where it's at sustained 2.3GB/s for 170GB file. It's true that the Evo's crapped out way earlier, so it does make me think that the SLC buffer implementation of the Sabrent Rocket is better than the 970 Evos. This transfer test in this post with the randomly generated text file above is not valid though, because my drive is at 100% read activity just trying to read it! That didn't happen with the dummy file that just contained zeros, so I think my testing in my previous posts is more valid for showing the write performance and SLC buffer limitations.

    EDIT: I also tested copy & paste of an 80GB Battlefield 1 game folder from one folder on the drive to another folder on the same drive:
    80GB Battlefield 1 folder transfer speed.jpg
    Again Read Activity is at 100% the whole time, so not totally valid, but the faster writes at the start do look like SLC buffer issues, but as you can see Read Activity is at 100% the whole time, so that could happen to be the bottleneck. It's not clear. This BF1 folder transfer test might be a good test to compare different NVMe drives anyway though, even if you're copying & pasting it from one folder to another on the same drive, as it seems to be a mix of both read & write performance. You guys could test that if you have BF1.

    Hmm, I'm not quite as impressed with my drive now, as I was kind of expecting to see the same stable 2.3GB/s I saw with the dummy file, but I've explained the issues with Read Activity being at 100%, but I reckon that BF1 folder copy & paste test within the same drive could be a good general indicator of NVMe drive performance (both read & write) - a means of comparison, because loads of people have BF1.

    (@tilleroftheearth , some new developments and contains some of the testing you referred to in your earlier post)

    EDIT#2: To sum up I suppose I'm just trying to see where the SLC cache limitations are with my Sabrent Rocket 1TB NVMe drive. Earlier testing from previous posts with a dummy file containing just zeros showed that there were effectively no SLC cache limitations - ie a constant 2.3GB/s write rate over a whole 170GB file, which destroys the Samsung 970 Evo drives when the same testing is done - just how valid is that testing in exposing SLC cache limitations, is it something to do with the fact that the dummy file just contains zeros that makes it work so well on the Sabrent vs the Samsung perhaps? My second lot of testing done in this post with files that don't contain just zeros shows slower transfer rates, but this looks like it's largely due to Read Rate bottlenecks rather than Write - so I don't think this testing does anything to expose SLC cache limitations on my drive, because of Read bottlenecks.
     
    Last edited: Jun 29, 2019
    jaybee83 likes this.
  8. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,126
    Messages:
    6,652
    Likes Received:
    5,798
    Trophy Points:
    681
    (Moderators, I'm happy for this SSD discussion to be moved to a different thread if necessary, maybe it can have a new thread created for it, but I'm equally as happy for the conversation to continue in here as long as no one is feeling uneasy about it).
     
  9. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,100
    Messages:
    11,564
    Likes Received:
    9,127
    Trophy Points:
    931
    hmmm im suspecting there are two factors at play here with regards to that initial 2.3 GB/s sustained copy that u got before.

    first one being the type of data being copied. the artificial file u created consisted of all zeroes, thus highly compressible data. rule of thumb is, the more compressible a data stream is, the easier it is to handle by ssd controllers and NAND flash and the faster the throughput.

    the second factor might be your overprovisioning. is definitely helps in endurance and performance, so i'm guessing a good apples to apples comparison would have to include those factors, as well.

    nevertheless, i would actually have expected to see some kind of performance degradation at some point, especially with a TLC based drive. so yeah, all in all quite the mystery those initial results you provided... lemme sleep on that and maybe ill come up with some more thoughts :)

    in any case, interesting data so far!

    Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk
     
    Last edited: Jun 30, 2019
    Rage Set and Robbo99999 like this.
  10. Rage Set

    Rage Set A Fusioner of Technologies

    Reputations:
    1,156
    Messages:
    1,354
    Likes Received:
    3,804
    Trophy Points:
    281
    :vbthumbsup:
     
    Mr. Fox likes this.
Loading...

Share This Page