AnandTech Storage Bench - The Destroyer

Our AnandTech Storage Bench tests are traces (recordings) of real-world IO patterns that are replayed onto the drives under test. The Destroyer is the longest and most difficult phase of our consumer SSD test suite. For more details, please see the overview of our 2021 Consumer SSD Benchmark Suite.

ATSB The Destroyer
Average Data Rate
Average Latency Average Read Latency Average Write Latency
99th Percentile Latency 99th Percentile Read Latency 99th Percentile Write Latency
Energy Usage

The SLC cache on the 2TB Intel 670p isn't large enough for The Destroyer to operate entirely within the cache, as we saw with the massive 8TB Sabrent Rocket Q. But the big SLC cache is still enough for the 670p to score very well overall on this test, clearly outperforming all the smaller entry-level NVMe SSDs we have tested, and more than a few mainstream and high-end models as well. The 670p's biggest weakness is with 99th percentile write latency, but even that score isn't problematic.

AnandTech Storage Bench - Heavy

The ATSB Heavy test is much shorter overall than The Destroyer, but is still fairly write-intensive. We run this test twice: first on a mostly-empty drive, and again on a completely full drive to show the worst-case performance.

ATSB Heavy
Average Data Rate
Average Latency Average Read Latency Average Write Latency
99th Percentile Latency 99th Percentile Read Latency 99th Percentile Write Latency
Energy Usage

As with The Destroyer, the empty-drive test run of the Heavy test shows that the 670p's performance can compete with good TLC drives. It's only on the full-drive test run that the QLC NAND starts to hold back the 670p. Even so, it fares better than almost all the competing entry-level drives and keeps the 99th percentile latencies down to reasonable values.

AnandTech Storage Bench - Light

The ATSB Light test represents ordinary everyday usage that doesn't put much strain on a SSD. Low queue depths, short bursts of IO and a short overall test duration mean this should be easy for any SSD. But running it a second time on a full drive shows how even storage-light workloads can be affected by SSD performance degradation.

ATSB Light
Average Data Rate
Average Latency Average Read Latency Average Write Latency
99th Percentile Latency 99th Percentile Read Latency 99th Percentile Write Latency
Energy Usage

On the Light test, the Intel 670p comes very close to matching the performance of mainstream TLC NVMe drives for both the empty and full drive test runs. Write latencies (average and 99th percentile) are still clearly higher than TLC drives, but not high enough to be a noticeable performance problem in storage-light real world usage. Power consumption is a bit on the high side, but that appears to be more due to the SSD controller than the downsides of QLC NAND.

PCMark 10 Storage Benchmarks

The PCMark 10 Storage benchmarks are IO trace based tests similar to our own ATSB tests. For more details, please see the overview of our 2021 Consumer SSD Benchmark Suite.

PCMark 10 Storage Traces
Full System Drive Overall Score Average Bandwidth Average Latency
Quick System Drive Overall Score Average Bandwidth Average Latency
Data Drive Overall Score Average Bandwidth Average Latency

Since we run the PCMark 10 Storage tests starting from an empty drive, the Intel SSD 670p is able to make full use of its large and very fast SLC cache. That puts it at the top of the charts for both the Quick System Drive and Full System Drive tests, and competitive with good TLC drives on the Data Drive test that is more geared toward sequential IO.

Introduction Synthetic Tests: Basic IO Patterns
POST A COMMENT

72 Comments

View All Comments

  • abufrejoval - Tuesday, March 2, 2021 - link

    I'd say a well used SATA SSD (unless the form factor is not negotiable).

    Booting is read-mostly and often not that I/O intensive after all.

    And then I generally try to avoid doing it, preferring systems that enable low-power idle or that will sleep. Pressing the power button first and then getting the coffee works wonders, too.
    Reply
  • Samus - Monday, March 1, 2021 - link

    I see Intel still refuses to give up their margins - no matter how uncompetitive or inferior the product may be.

    Looks like a good drive but the pricing is around 50% too high.
    Reply
  • Glock24 - Monday, March 1, 2021 - link

    I've never, NEVER seen any QLC drive worth buying. The pricing is always really bad in relation to brand name TLC drives. Even if they were cheaper, it's not worth because of all the drawbacks. Reply
  • Byte - Tuesday, March 2, 2021 - link

    where does SSDs have to go from here? OLC? Would that be even possible? Or do we have to wait for the next breakthrough. Reply
  • Zizy - Tuesday, March 2, 2021 - link

    PLC - penta-level cell. Each cell keeps 5 values -> 32 voltage levels are required to read that (2^N). It is possible, but it will take a while. Reply
  • dragosmp - Tuesday, March 2, 2021 - link

    The gain in flash performance is impressive and worth keeping in mind for when Intel will drop the price to something reasonable.

    I have used a 1TB P1 in my gaming rig for close to 2 years now and it's solid. Worth keeping in mind most SSDs nowadays come with RAM caching software which hides some of the el-cheapo's disadvantages. Not about to say the P1 is a speed demon, but it doesn't have to be if it has 4GB of RAM buffer. I'd like this drive for the write endurance and 5 years of warranty though, which is a deffinite plus; when it gets below 200$/2TB, preferably below 150 on offer.
    Reply
  • Wereweeb - Tuesday, March 2, 2021 - link

    First of all: remember that TLC, QLC, etc... are not atemporal constants. They're different ways to build NAND Flash storage. E.g.: Planar TLC is comparable to 3D QLC in a lot of ways.

    And QLC/PLC are not inevitable. They could tweak other properties of 3D NAND to enable progress. If they ever find a way to substantially shrink 3D NAND, they'd have to go back to TLC, because otherwise QLC would behave like what we imagine PLC to be.

    Second: current 3D QLC is simply good enough for 95% of consumers. Look at these numbers. At the right price, there would be no reason not to buy this if you don't have some kind of professional application which requires consistently high performance storage.

    Third: Yeah, please don't buy SSD's from Intel anymore. There's absolutely no benefit to QLC if it isn't substantially cheaper. A Dramless TLC is also good enough for 95% of consumers and it likely won't commit sudoku in a couple of years.
    Reply
  • Wereweeb - Tuesday, March 2, 2021 - link

    Sorry, I meant "different properties to tweak in building NAND storage" Reply
  • Spunjji - Thursday, March 4, 2021 - link

    100% behind this comment. I've considered buying Intel's QLC SSDs when I've seen them on sale for substantially less than TLC alternatives, but at retail price, they're a joke. Reply
  • HVAC - Monday, March 8, 2021 - link

    But my SSDs are tasked exclusively to solve sudoku! What am I supposed to do after a couple of years? Use my worn out NVMe sticks to spread margarine on my pretzels?

    It's distressing enough to make one want to commit sudoku!
    Reply

Log in

Don't have an account? Sign up now