Advanced Synthetic Tests

Our benchmark suite includes a variety of tests that are less about replicating any real-world IO patterns, and more about exposing the inner workings of a drive with narrowly-focused tests. Many of these tests will show exaggerated differences between drives, and for the most part that should not be taken as a sign that one drive will be drastically faster for real-world usage. These tests are about satisfying curiosity, and are not good measures of overall drive performance. For more details, please see the overview of our 2021 Consumer SSD Benchmark Suite.

Whole-Drive Sequential Fill

Pass 1
Pass 2

Some of our other tests have shown a few signs that the 870 EVO's write performance can drop when the SLC cache runs out, but this straightforward sequential write pass over the entire drive doesn't reveal any such behavior. The 870 EVO's sequential write performance is extremely consistent, even on the second write pass.

Sustained 128kB Sequential Write (Power Efficiency)
Average Throughput for last 16 GB Overall Average Throughput

Due to the excellent performance consistency, the Samsung 870 EVOs edge out the other SATA drives with marginally higher average sequential write speeds. The entry-level NVMe drives end up much worse off than the mainstream SATA drives once their caches run out, but the more mainstream NVMe drive blows them all away.

 

Working Set Size

As expected, the Samsung 870 EVO's random read performance shows basically no variation across a range of working set sizes, and that read performance is at least a little bit faster than any of the other SATA drives or the entry-level NVMe drives.

 

Performance vs Block Sizes

Random Read
Random Write
Sequential Read
Sequential Write

There are no particular surprises in how the Samsung 870 EVO handles IOs of different block sizes. Unlike some drives, it has no trouble with sub-4kB IOs. It offers moderate improvements over the 860 EVO for mid-sized random reads (up to about 128kB). The one negative is that for writes we again see more inconsistency from the 870 EVO than the 860 EVO when testing an 80% full drive. The simple whole-drive sequential write test may not have been able to reveal any SLC caching troubles, but it does seem clear that the caching behavior has some performance regressions for more complicated workloads on a drive that's more well-used—though it's still unlikely to matter for any typical real-world consumer workload.

Synthetic Tests: Basic IO Patterns Mixed IO Performance and Idle Power Management
POST A COMMENT

139 Comments

View All Comments

  • Jorgp2 - Wednesday, February 17, 2021 - link

    Nah, it all comes down to EFI vs MBR.

    MBR systems just look for a boot partition and boot it, EFI actually stores the location on the motherboard flash.
    Reply
  • WaltC - Wednesday, February 17, 2021 - link

    Agreed...;) I might say "UEFI vs. Legacy," though. And much of it has to do with the knowledge, experience, and skill of the computer operator...! No question about that. (It's actually MBR vs. GPT--GPT is better. All of my SATA HDDs are formatted GPT, etc. I have no MBR-formatted drives.)

    The list of NVMe drives here is curious. Where is the 980 Pro from Samsung?--been selling for a while now--one of several PCIe4 NVMe drives available. Makes me question when this article was actually written....? It seems incomplete or out of date. Samsung is a chip company no longer in the business of making platter drives (last I looked, anyway-do they still sell the Spinpoints?), so it's natural for them to sell SSDs of varying types, sizes, and prices, imo.
    Reply
  • Billy Tallis - Wednesday, February 17, 2021 - link

    I deliberately chose not to include 980 PRO results in the graphs for this review, because that's a silly comparison to make against a SATA drive. But if you really care, you can use Bench: https://www.anandtech.com/bench/product/2724?vs=27... Reply
  • Gigaplex - Wednesday, February 17, 2021 - link

    A half-decent UEFI implementation will enumerate the bootloaders (or fall back to the default /EFI/Boot/bootx64.efi path) if the drive isn't configured in the motherboard. Reply
  • Billy Tallis - Thursday, February 18, 2021 - link

    Unfortunately, "half-decent" can't be taken for granted. The ASRock motherboard in the new SSD testbed won't look in the standard path for a bootloader, but it will happily boot any Windows bootloader it finds. Reply
  • Duncan Macdonald - Wednesday, February 17, 2021 - link

    SATA may be declining but is far from dead - many motherboards (and laptops) only have one NVMe slot - adding a SATA drive is far easier than replacing the NVMe drive when it is the system drive. Also NVMe drives bigger than 4TB are rare and expensive so anyone needing large storage capacity (over 4TB) has a choice between SATA or expensive NVMe. Reply
  • npz - Wednesday, February 17, 2021 - link

    My thoughts too. Until I can get a 4tb or more nvne m.2 drive priced the same and also not have it throttle under sustained i/o I'll be sticking with mix of mechanical and ssd sata drives and u..2 drines and m.2 placed into u.2 adapters for 2.5 bay cooling for those few hot nvme drives I can't place a tall heatsink over Reply
  • Gigaplex - Wednesday, February 17, 2021 - link

    If you're prepared to suffer from mechanical performance limitations, then you shouldn't be worried about sustained IO throttling. Reply
  • npz - Thursday, February 18, 2021 - link

    As I said, I have a *mix* of drives for different workloads. If I'm doing concurrent read and write from a source drive to a destination drive for large video files I'm not doing that on the mechanical drives and doing that on m.2 drives without proper cooling will throttle. Reply
  • Oxford Guy - Friday, February 19, 2021 - link

    Some motherboard brands also don't know how to design a motherboard properly, like Gigabyte.

    Drives like the Inland Performance Plus (Phison) don't fit in boards like the Z390 UD.
    Reply

Log in

Don't have an account? Sign up now