Random Read Performance

Our first test of random read performance uses very short bursts of operations issued one at a time with no queuing. The drives are given enough idle time between bursts to yield an overall duty cycle of 20%, so thermal throttling is impossible. Each burst consists of a total of 32MB of 4kB random reads, from a 16GB span of the disk. The total data read is 1GB.

Burst 4kB Random Read (Queue Depth 1)

The burst random read performance of the Samsung 970 EVO is the best they've ever delivered from TLC NAND flash memory, but the Intel SSD 760p is a few percent faster still.

Our sustained random read performance is similar to the random read test from our 2015 test suite: queue depths from 1 to 32 are tested, and the average performance and power efficiency across QD1, QD2 and QD4 are reported as the primary scores. Each queue depth is tested for one minute or 32GB of data transferred, whichever is shorter. After each queue depth is tested, the drive is given up to one minute to cool off so that the higher queue depths are unlikely to be affected by accumulated heat build-up. The individual read operations are again 4kB, and cover a 64GB span of the drive.

Sustained 4kB Random Read

On the longer random read test, the Samsung 970 EVO proves to be the fastest TLC-based drive, but Samsung's MLC-based drives offer up to 20% higher performance.

Sustained 4kB Random Read (Power Efficiency)
Power Efficiency in MB/s/W Average Power in W

The Samsung 970 EVO and its OEM sibling PM981 have the worst power efficiency of any recent high-end SSD during the random read test. The 970 EVO is drawing over 2.5W while Samsung's previous generation high end drives averaged less than 2W for very similar performance.

The performance scaling of the 970 EVO is almost identical to that of the 960 EVO, but the 970 EVO draws more power throughout the random read test.

Random Write Performance

Our test of random write burst performance is structured similarly to the random read burst test, but each burst is only 4MB and the total test length is 128MB. The 4kB random write operations are distributed over a 16GB span of the drive, and the operations are issued one at a time with no queuing.

Burst 4kB Random Write (Queue Depth 1)

The burst random write performance from the Samsung 970 EVO is disappointing compared to the PM981, especially for the 1TB 970 EVO. Meanwhile, recent Intel and WD drives have been raising the bar with very fast SLC write caches.

As with the sustained random read test, our sustained 4kB random write test runs for up to one minute or 32GB per queue depth, covering a 64GB span of the drive and giving the drive up to 1 minute of idle time between queue depths to allow for write caches to be flushed and for the drive to cool down.

Sustained 4kB Random Write

On the longer random write test, the 1TB PM981 provided top-tier performance, but the 1TB 970 EVO is about 12% slower, putting it on par with the previous generation from Samsung. The 500GB 970 EVO is also slightly slower than its PM981 counterpart.

Sustained 4kB Random Write (Power Efficiency)
Power Efficiency in MB/s/W Average Power in W

Power efficiency has also regressed for the 970 EVO on the random write test, leaving it well below the standard set by the WD Black and the slower but similarly efficient Toshiba XG5.

The random write performance of the 1TB 970 EVO tops out at just over 1.5 GB/s at queue depths of 8 and higher. The 500GB 970 EVO starts running out of SLC cache and showing inconsistent performance past QD4. The 1TB PM981 was able to ramp up performance much faster than the 970 EVO and hit a maximum of about 1.8GB/s before running out of SLC cache near the end of the test. The 512GB PM981 behaved very similarly to the 500GB 970 EVO.

AnandTech Storage Bench - Light Sequential Performance
Comments Locked

68 Comments

View All Comments

  • qlum - Tuesday, October 16, 2018 - link

    Just as a reminder how the argument against the ssd can change overtime:
    Right now the prices at least here in the netherlands are as follows:

    Crucial MX500 (SATA) €160
    HP EX920 (NVMe PCIe x4) €287
    Intel 760p (NVMe PCIe x4 ) €289
    WD Black (NVMe PCIe x4) €312
    Samsung 970 EVO (NVMe PCIe x4) €269
    Samsung 960 PRO €374

    Suddenly the 970 evo is the cheapest of the bunch this makes its value a lot better

    Of course there are still cheaper nvme ssds such as the intel 660p but at the tb mark its one of the cheapest nvme ssd's
  • modeonoff - Tuesday, April 24, 2018 - link

    Isn't nvme m.2 SSD performance affected by Meltdown/Spectre patches?
  • Billy Tallis - Tuesday, April 24, 2018 - link

    Yes, because storage benchmarks make system calls more frequently than almost anything else. Once the updates have been applied to the testbed, I'll be re-testing everything for future reviews. This will take a while, so I've waited until I have several reviews worth of testing completed that can fill the gap before I have new results for a new drive and the older drives it needs to be compared against.

    My preliminary tests of the impact of the patches show that while the scores themselves are affected, the rankings of drives aren't, so the current measurements are still useful for judging which drives are best.
  • Reppiks - Tuesday, April 24, 2018 - link

    Could be nice to see AMD vs Intel post patches as it shouldn't affect AMD as much?
  • Infy2 - Tuesday, April 24, 2018 - link

    Nvme and Sata controllers are made by AsMedia on AMD's AM4 mother boards. Sadly they are somewhat slower than Intel's controllers. Even after Spectre and Meltdown patches Intel is still king of storage performance.
  • Tamz_msc - Tuesday, April 24, 2018 - link

    Ryzen CPUs have a dedicated PCI-E x4 link for nvme drives which bypasses the chipset.
  • bernstein - Tuesday, April 24, 2018 - link

    nvme is just PCIe x4 + software... that's why a passive pcie x4 to m.2 works. and why a nvme m.2 ssd should work with reduced speed over PCIe x1 or PCIe x2. the same goes for PCIe 2.0 links... combine these and you get a working passive mPCIe to M.2 adapter.
  • willis936 - Tuesday, April 24, 2018 - link

    NVMe controllers are made by Intel, AMD, and Microsoft (and whatever analog set of companies for mobile) because it's just a software stack that runs on CPUs.
  • Kwarkon - Wednesday, April 25, 2018 - link

    Close but not exactly. You mean drivers.
  • HStewart - Tuesday, April 24, 2018 - link

    I personally think people are making a bigger deal of this Methdown/Spectre stuff than it worth it.

    Yes performance is one area - but there are other reasons why people purchase a product.

    Especially in heavy graphics or in this case storage usage - these patches should not have no minimal effect.

    To the average customer - the effect is not notice - how much will they notice a 5% or leas slow down in cpu speed. But change a hard drive to one of these SSD's would be a significant improvement

Log in

Don't have an account? Sign up now