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 average data rate from the Crucial P1 on The Destroyer is comparable to other entry-level NVMe drives like the Phison E8-based Kingston A1000 and the Intel 660p. The P1 also roughly matches the average data rate of the Crucial MX500 SATA SSD, while several high-end NVMe drives deliver more than twice the performance.

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

While the average data rate from the Crucial P1 may have been similar to the MX500, the average latency is about twice that of the MX500, and slightly higher than the Intel 660p. The situation is better for the 99th percentile latency, where the Crucial P1 comes close to the MX500 and shows half the 99th percentile latency of the Intel 660p.

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

The average read and write latencies for the Crucial P1 on The Destroyer are both slightly worse than what the Intel 660p provides, and the Crucial MX500 provides a much better average write latency.

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

The 99th percentile read latency from the Crucial P1 is very good, competitive with several high-end NVMe SSDs. However, the 99th percentile write latency is quite poor compared to almost any other NVMe SSD or mainstream SATA SSD. This is a huge difference in behavior compared to the Intel 660p. The Crucial P1 is optimized much more for reliable read latency, at significant cost to worst-case write latency under heavy workloads.

ATSB - The Destroyer (Power)

The Crucial P1 requires significantly more energy to complete The Destroyer than the Intel 660p, despite the near-identical hardware and very similar overall performance numbers. The Crucial P1's relatively poor efficiency on this test isn't a serious issue given that the drive is intended for less demanding use cases, but combined with the high 99th percentile write latency this points to the P1 possibly experiencing higher write amplification on The Destroyer than the Intel 660p experiences.

Introduction AnandTech Storage Bench - Heavy
Comments Locked


View All Comments

  • DanNeely - Thursday, November 8, 2018 - link

    When DDR2 went mainstream they stopped making DDR1 dimms. The dimms you could still find for sale a few years later were old ones where you were paying not just the original cost of making them, but the cost of keeping them in a warehouse for several years before you bought them. Individual ram chips continued to be made for a while longer on legacy processes for embedded use but because the same old mature processes were still being used there was no scope for newer tech allowing cost cutting, and lower volumes meant loss of scale savings meaning that the embedded world also had to pay more until they upgraded to new standards.
  • Oxford Guy - Thursday, November 8, 2018 - link

    The point was:

    "QLC may lead to higher TLC prices, if TLC volume goes down and/or gets positioned as a more premium product as manufacturers try to sell us QLC."

    Stopping production leads to a volume drop, eh?
  • romrunning - Thursday, November 8, 2018 - link

    "There is a low-end NVMe market segment with numerous options, but they are all struggling under the pressure from more competitively priced high-end NVMe SSDs."

    I really wish all NVMe drives kept a higher base performance level. QLC should have died on the vine. I get the technical advances, but I prefer tech advances increase performance, not ones that are worse than their predecessor. The price savings, when it's actually there, isn't worth the trade-offs.
  • Flunk - Thursday, November 8, 2018 - link

    In a year or two there are going to be QLC drives faster than today's TLC drives. it just takes time to develop a new technology.
  • Oxford Guy - Thursday, November 8, 2018 - link

    Faster to decay, certainly.

    As I understand it, it's impossible, due to physics, to make QLC faster than TLC, just as it's impossible to make TLC faster than MLC. Just as it's impossible to make MLC faster than SLC.

    Workarounds to mask the deficiencies aren't the same thing. The only benefit to going beyond SLC is density, as I understand it.
  • Billy Tallis - Thursday, November 8, 2018 - link

    Other things being equal, MLC is faster than TLC and so on. But NAND flash memory has been evolving in ways other than changing the number of bits stored per cell. Micron's 64L TLC is faster than their 32L MLC, not just denser and cheaper. I don't think their 96L or 128L QLC will end up being faster than 64L TLC, but I do think it will be faster than their 32L or 16nm planar TLC. (There are some ways in which increased layer count can hurt performance, but in general those effects have been offset by other performance increases.)
  • Oxford Guy - Thursday, November 8, 2018 - link

    "Other things being equal, MLC is faster than TLC and so on"

    So, other than density, there is no benefit to going beyond SLC, correct?
  • Billy Tallis - Thursday, November 8, 2018 - link

    Pretty much. If you can afford to pay for SLC and a controller with enough channels and chip enable lines, then you could have a very nice SSD for a very unreasonable price. When you're constrained to a SATA interface there's no reason not to store at least three bits per cell, and even for enterprise NVMe SSDs there are only a few workloads where the higher performance of SLC is cost-effective.
  • Great_Scott - Monday, November 12, 2018 - link

    They should drop the SLC emulation and just sell the drive as an SLC drive. Sure, there may be some performance left on the table due to the limits of the NVME interface, but the longevity would be hugely attractive to some users.

    They'd make more money too, since they could better justify higher costs that way. In fact, with modern Flash they might be able to get much the same benefit from MLC organization and have roughly half the drive space instead of 25%.
  • Lolimaster - Friday, November 9, 2018 - link

    Do not mix better algorithms of the simulated SLC cache and dram with actual "performance", start crushing their simulated cache and the TLC goes to trash.

Log in

Don't have an account? Sign up now