Mixed Random Performance

Our test of mixed random reads and writes covers mixes varying from pure reads to pure writes at 10% increments. Each mix is tested for up to 1 minute or 32GB of data transferred. The test is conducted with a queue depth of 4, and is limited to a 64GB span of the drive. In between each mix, the drive is given idle time of up to one minute so that the overall duty cycle is 50%.

Mixed 4kB Random Read/Write

The mixed random I/O performance of the Samsung PM981 is a big improvement over last generation's 960 EVO. The 1TB PM981 beats out even the MLC-based 960 PRO, while the smaller 512GB PM981 is a bit slower than the 960 PRO of the same size.

As the proportion of writes in the mixed workload increases, the PM981 steadily gains performance, pulling further and further ahead of the 960 EVO. The 512GB PM981's main weakness is that its performance doesn't hit quit as high a peak during the final phases of the test when the workload is almost entirely random writes.

Mixed Sequential Performance

Our test of mixed sequential reads and writes differs from the mixed random I/O test by performing 128kB sequential accesses rather than 4kB accesses at random locations, and the sequential test is conducted at queue depth 1. The range of mixes tested is the same, and the timing and limits on data transfers are also the same as above.

Mixed 128kB Sequential Read/Write

The 512GB PM981 matches the mixed sequential performance of the MLC-based 512GB 960 PRO, while the 1TB PM981 is substantially faster than the 960 PRO or any other flash-based SSD.

The Samsung 960 PRO 1TB outperforms the 1TB PM981 during the early read-heavy phases of the mixed sequential test, but then its performance drops off precipitously while the PM981 retains its performance until later in the test. The 512GB PM981 averages almost exactly the same performance as the 512GB 960 PRO, but with substantial differences in the details: the 960 PRO is faster at either end of the test, but the PM981 has a significant advantage for more even mixes of reads and writes.

Sequential Performance Conclusions
Comments Locked

53 Comments

View All Comments

  • peevee - Thursday, November 30, 2017 - link

    The most important parameter is the sustained random read, when the user have to actually wait (buffered writes let you continue working right away and it is almost impossible to overflow write caches during normal desktop usage).
    And even SSDs continue to suck in this parameter. 60MB/s? Booo...
    Although testing on 4k random is too strict, NTFS runs are usually 16 clusters (64k).
  • wyewye - Thursday, November 30, 2017 - link

    This is why I stopped coming to Anandtech daily: you keep excluding top offers from Intel in your SSD benchmarks.

    Samsung shills!
  • ddriver - Thursday, November 30, 2017 - link

    LOL, second funniest thing I heard this week
  • ddrіver - Thursday, November 30, 2017 - link

    The first one was actually a joke with a priest and a rabbi... Can't really remember the punchline now.
  • lilmoe - Thursday, November 30, 2017 - link

    Thank you, competing OEMs. Thanks to you, Samsung isn't even trying anymore. Just when I thought they'd introduce 64 layer SLC, they decide to go full TLC, because why try harder? Screw you too, Samsung.
  • ddrіver - Thursday, November 30, 2017 - link

    Exactly my point. It's not so hard for them to go to 64 or 128 or even 256 layer SLC, with even an 8TB SSD. That should be in the range of $500-$600 to be competitive. Instead they choose to deliver a drive that simply doesn't massively improve on every single data point relative to the old generation. It might be good for 95% of consumers but they don't even think of us, professionals.

    Just downloading a game nowadays takes 50-100GB of SSD writes. At this rate who know how long I'm going to be able to use this kind of SSD. Greedy people selling to sheeple. Wake up!
  • zodiacfml - Friday, December 1, 2017 - link

    Samsung, like Intel, has no competition taking their sweet time with each iteration.
  • AnnonymousCoward - Saturday, December 2, 2017 - link

    In many cases Samsung costs more and has less endurance; the competition is better.
  • melgross - Friday, December 1, 2017 - link

    It’s good to see that manufacturers are so far in advance of where they already are:

    “Other M.2 PCIe SSD vendors have used that tactic and many have also released drives with more substantial heatspreaders or heatsinks in the future.”
  • trumanhw - Sunday, December 31, 2017 - link

    1TB 960 Pro looks like it's the shiznit. By far my choice... cost per dollar, and as they said, impunity to the negatives seen in other sizes and models. Where it's not the first, it's so close in the other positions as to require measurement to verify. As always, thank you anand -- you guys rock.

Log in

Don't have an account? Sign up now