AnandTech Storage Bench - Light

Our Light storage test has relatively more sequential accesses and lower queue depths than The Destroyer or the Heavy test, and it's by far the shortest test overall. It's based largely on applications that aren't highly dependent on storage performance, so this is a test more of application launch times and file load times. This test can be seen as the sum of all the little delays in daily usage, but with the idle times trimmed to 25ms it takes less than half an hour to run. Details of the Light test can be found here. As with the ATSB Heavy test, this test is run with the drive both freshly erased and empty, and after filling the drive with sequential writes.

ATSB - Light (Data Rate)

The Samsung 860 QVO has no trouble with the Light test when it is run on an empty drive, and the full-drive performance loss is not too bad: the 1TB 860 QVO remains ahead of the DRAMless TLC drive even when the drives are full.

ATSB - Light (Average Latency)ATSB - Light (99th Percentile Latency)

The average and 99th percentile latency scores from the 860 QVO are no problem when the test is run on a full drive. They're substantially higher when the drives are full, but the latency is better-controlled than on the Intel/Micron QLC drives.

ATSB - Light (Average Read Latency)ATSB - Light (Average Write Latency)

The average read and write latency scores from the 860 QVO are clearly different from the TLC drives for the full-drive test runs, but they don't stand out as significantly worse than what we've seen from some of the slower TLC drives.

ATSB - Light (99th Percentile Read Latency)ATSB - Light (99th Percentile Write Latency)

The 99th percentile read latency on the 860 QVO is a sore spot when the drive is full, but the 99th percentile write latency doesn't get too far out of control, especially compared to the other two QLC drives.

ATSB - Light (Power)

All of the QLC drives use more energy than the TLC drives during the Light test, and especially when the drives are full and have more background work to do.

AnandTech Storage Bench - Heavy Random Performance


View All Comments

  • Impulses - Thursday, November 29, 2018 - link

    Bleh, googling revealed that WAS on Amazon... It never triggered my price alerts, hrm, even tho it's definitely showing on Camelcamel's price history now, weird. Maybe I glossed over it entirely, oh well. Reply
  • The_Assimilator - Thursday, November 29, 2018 - link Reply
  • hojnikb - Tuesday, November 27, 2018 - link


    Can you do a data retention test ? Like writing the drive with data (H2testw seems to be the pick of the bunch) and disconnecting it. I'm interested how data retention holds up over time.
  • JoeyJoJo123 - Tuesday, November 27, 2018 - link

    Don't use SSDs for cold storage. Testing this is absurdly stupid and a waste of a reviewers time.

    Disconnect drive for 10 days: Replug it in and verify data's still all there. Yep.
    Disconnect drive for 15 days: Replug it in and verify data's still all there. Yep.
    Disconnect drive for 20 days: Replug it in and verify data's still all there. Yep.
    (We're already at 45 days, or 1.5 months, and chances are the data's still all there, are you getting how and why this kind of testing is stupid?)
    By the time you get to sufficiently long enough time periods to see some change in data you'd have wasted over a year trying to find a time period (within 5 days) where you know the drive will start losing data. This is just as absurd as asking reviewers to test NAND flash endurance by hammering the drive 24/7 for years until it dies. By the time it DOES die something better will have already arrived on the market. And disregarding all that, testing with a sample size of ONE is not indicative of any relevant performance characteristics for your ONE drive.

    If you want to use cold storage backups, either use a mechanical hard drive, a tape drive, or invest in cloud storage and encrypt your data before uploading it.

    Testing the cold storage capabilities of a sample size of ONE QLC SSD does nothing but prove that it's a less satisfactory use case for the technology than using a mechanical hard drive.
  • HollyDOL - Tuesday, November 27, 2018 - link

    I _think_ you can find Google or Facebook statistics on their SSDs, error rate etc. in representative volumes. But ofc it is enterprise grade hardware. Reply
  • hojnikb - Wednesday, November 28, 2018 - link

    Not for QLC. Reply
  • hojnikb - Wednesday, November 28, 2018 - link

    You're not thinking very far. As QLC is very cheap, it will be used in likes of flash drives, SD cards and portable SSDs. With these, it's not expected for them to be powerred at all time and in the case of sd cards and flash drives, it's not likely controller would do any kind of data rewritting.

    So this is very much a relavant test, that might give us ideas on how it performs. And no, you don't have to wait a year to see something; data curruption can be presented a lot sooner if from of read error and slower read speed in general. A more extreme approach is to heat the drive, which accelarates the process.
  • eddieobscurant - Tuesday, November 27, 2018 - link

    Horrible performance by samsung 64 layer qlc nand. Way slower than intel's/micron's . Especially the 4k random reads are slower than my first ssd 10 years ago, the intel x-25m.

    In order for samsung 860 qvo to make sense, it should be priced below 0.10$/gb
  • stargazera5 - Tuesday, November 27, 2018 - link

    Great, now that we have QLC down, we can move on to PLC (5 bits per cell) and 1 drive write per week. Reply
  • CheapSushi - Wednesday, November 28, 2018 - link

    And from the current trend, it still won't be much cheaper. =P Reply

Log in

Don't have an account? Sign up now