AnandTech Storage Bench - The Destroyer

The Destroyer is an extremely long test replicating the access patterns of very IO-intensive desktop usage. A detailed breakdown can be found in this article. Like real-world usage, the drives do get the occasional break that allows for some background garbage collection and flushing caches, but those idle times are limited to 25ms so that it doesn't take all week to run the test. These AnandTech Storage Bench (ATSB) tests do not involve running the actual applications that generated the workloads, so the scores are relatively insensitive to changes in CPU performance and RAM from our new testbed, but the jump to a newer version of Windows and the newer storage drivers can have an impact.

We quantify performance on this test by reporting the drive's average data throughput, the average latency of the I/O operations, and the total energy used by the drive over the course of the test.

ATSB - The Destroyer (Data Rate)

The NX500's average data rate on The Destroyer is clearly faster than the other Phison E7 drives, but still far slower than competing MLC NVMe SSDs. The Plextor M8Pe and OCZ RD400 offer much higher performance from the same 15nm MLC NAND.

ATSB - The Destroyer (Average Latency)ATSB - The Destroyer (99th Percentile Latency)

The NX500 offers the best average latency on The Destroyer among the Phison E7 drives. The 99th percentile latencies are quite different, with the Zotac SONIX outperforming the Corsair NX500, while the Patriot Hellfire is clearly having problems keeping latency under control.

ATSB - The Destroyer (Average Read Latency)ATSB - The Destroyer (Average Write Latency)

Splitting the average latency by reads and writes doesn't reveal any surprises among the Phison drives, but there are a few differences in the rankings of the other drives. The Intel 600p stands out as the only drive that has a real problem with write latency even in the average case.

ATSB - The Destroyer (99th Percentile Read Latency)ATSB - The Destroyer (99th Percentile Write Latency)

The 99th percentile latencies show interesting differences between the Phison drives in how they handle reads and writes. The Zotac SONIX has the best 99th percentile read latency among the Phison E7 drives and places pretty well among the overall field of competitors. The Corsair Neutron NX500 is substantially slower, while the Patriot Hellfire's 99th percentile read latency is worse than any other NVMe drive on this chart.

The 99th percentile write latencies are all rather slow on the Phison E7 drives, but the Corsair NX500 is the best of the three at keeping write latency low. This is to be expected given the greater spare area the NX500 has to work with, but the impact of that much extra overprovisioning on write QoS ought to be more significant.

ATSB - The Destroyer (Power)

The Corsair Neutron NX500 and Zotac SONIX are at a bit of a disadvantage compared to the M.2 drives here since the NX500 and SONIX add-in cards draw power from the 12V supply and thus include more voltage regulation circuitry than the M.2 SSDs. The NX500 is further disadvantaged by having more DRAM to power. However, even the Patriot Hellfire M.2 can't come close to the 3D NAND SSDs from Samsung and Toshiba. Surprisingly, the most efficient planar NAND SSD in this bunch is the Plextor M8PeY, which was flashing its red LEDs during the entire test.

Introduction AnandTech Storage Bench - Heavy
Comments Locked

45 Comments

View All Comments

  • Kristian Vättö - Wednesday, August 16, 2017 - link

    There are numerous 8-lane enterprise SSDs already.
  • hlm - Wednesday, August 16, 2017 - link

    e.g. HGST FlashMAX III and HGST Ultrastar SN260 products are eight-lane devices.
  • The_Assimilator - Wednesday, August 16, 2017 - link

    Hey look, another SSD that has no reason whatsoever to exist!

    I don't understand why manufacturers don't, y'know, try to COMPETE with Samsung instead of re-re-releasing the same old, tired, slow controllers with slightly different but ultimately insignificant spins on them.
  • DanNeely - Wednesday, August 16, 2017 - link

    Because unless you have a billion dollars to spend and a few years to wait, you can't create your own controller. That means almost all of the other companies selling drives have to pick and choose between a handful of controllers made by Phison/etc. Until they recover from Samsung's blind siding them and design new higher performing architectures from the ground up none of them have anything in the same performance class. If what happened at the start of the market when Intel's controllers were unbeatable is any indication we should hopefully have competitive designs available in another year or so.
  • FunBunny2 - Wednesday, August 16, 2017 - link

    -- Because unless you have a billion dollars to spend and a few years to wait, you can't create your own controller.

    well, isn't a controller an implementation of physics and math? which is to say, unless something new happens with NAND chips (not just node size or xLC), may haps we've reached the one-true-answer to the controller problem? may be there's just no more there, there.
  • Samus - Wednesday, August 16, 2017 - link

    Wow. That was disappointing.
  • RaistlinZ - Wednesday, August 16, 2017 - link

    Current Newegg Prices:

    1. 500GB Samsung 960 Pro = $299.99
    2. 1TB Samsung 960 Pro = 600.82

    The NX500 has no reason to exist. The price needs to be cut in half to make it even REMOTELY attractive.
  • alpha754293 - Wednesday, August 16, 2017 - link

    I'm surprised you didn't bother comparing it against the Intel 750 Series 400 GB PCIe NVMe SSD.
  • Billy Tallis - Thursday, August 17, 2017 - link

    I had originally planned to include the 400GB 750, but some of the results from it looked funny and I decided it wasn't worth postponing the review for several days to re-test the 750. That drive's a pain to test, because I have to run each test twice in order to record the power on both the 3.3V and 12V lines, and the performance has to match between the two runs for the results to be valid.
  • alpha754293 - Friday, August 18, 2017 - link

    Depending on how you want to tackle/handle it.

    There are statistical methods available out there that even with noisy data (e.g. high standard deviations) that you can still use it to process data that might otherwise not make sense at first glance, on the surface.

    Course, that would also mean that care would need to be taking so that the tests in and of itself are repeatable.

    I only mention it because I would be VERY interested to see how this compared to the Intel 750 series.

    Thanks.

Log in

Don't have an account? Sign up now