WD Black 3D NVMe SSD Review: WD Debuts Its New Controller

What is the secret to success? In the SSD business, it's vertical integration. That means designing all of the important components so they can work together to deliver the best possible performance and reliability at a low price point. In the past, only Samsung, Toshiba, and Intel produced their own flash and SSD controllers. Now, thanks to subsidiary SanDisk, Western Digital joins the club.

WD's new Black 3D NVMe SSD comes equipped with a new proprietary SSD controller that serves as the foundation of a far-reaching initiative, so several more enhancements will come in time. The initial launch includes two controllers: the high-performance version we will see today and a power efficient DRAM-less version for the embedded market. WD introduced the latter on the PC SN520 SSD just a few weeks ago.

The new controller comes at an odd time. Any new controller is a welcome addition to the market, but Western Digital could have used the controller with its previous-generation WD Black PCIe SSD. Unfortunately, that SSD suffered from poor performance thanks to a lackluster Marvell controller. WD had to choose between two evils during the design process, though. There simply wasn't a high-performance third-party controller available to build an NVMe SSD, so the company had to choose between dual-core Marvell and SMI controllers. Neither controller produces enthusiast-class performance.

Now, after three years in development, WD's new proprietary SSD controller is ready to take the company's enthusiast-class SSDs to the next level.

Specifications

The WD Black 3D NVMe SSD, which we'll refer to as the Black, comes to market in three capacities that span from 1TB to 250GB. The SanDisk Extreme Pro NVMe SSD, which has identical hardware, will come to market with 1TB and 500GB capacities. The Extreme Pro is designed for "creatives" while the Black is for gamers, enthusiasts, and power users.

On paper, the new Black delivers up to 3,400/2,800 MB/s of sequential read/write throughput thanks to the PCI Express 3.0 x4 interface paired with the latest NVMe 1.3 protocol. As expected, the two smaller drives lose some sequential write performance. The 1TB model pushes up to 500,000/400,000 random read/write IOPS, but we also see reduced random performance with the smaller models. Unlike some of its competitors, Western Digital advertises the performance for all three models separately.

WD paired the new controller with its 64-layer BICS TLC NAND that has already proven itself in other products. We've detailed BiCS flash, which is jointly produced by WD and Toshiba, several times already. Toshiba sent us the XG5 NVMe SSD with BICS flash in 2017.

Features

WD's new NVMe controller is the real news today. It should power WD's new products for several years, including QLC SSDs that could debut as early as 2019. WD chose a 28nm tri-core design to target the same market Samsung addresses with the 960 EVO. Samsung uses a five-core ARM design for the 960 Pro and various enterprise variants.

WD designed its new controller to allow some data to pass through the processor without consuming clock cycles for data processing. This technique reduces temperature and power consumption while increasing performance.

SanDisk also updated its nCache to version 3. This is the company's proprietary intelligent cache that assigns a portion of the NAND to operate as SLC, which boosts performance. The drives have a fixed amount of SLC-programmed flash, so smaller drives have less cache.

The Black's SSD controller passes most of the incoming write operations directly to the SLC cache, but that isn't always ideal. Some SSDs force all the incoming data to pass through the cache before it transfers the data to the primary TLC storage pool. That can slow transfer speeds if the SLC cache is already saturated with incoming data. To circumvent this problem, WD's controller can write directly to the TLC flash if needed. This eliminates the process of flushing the SLC cache before accepting new data. In our opinion, this is more favorable. Our internal tests have proven that bypassing the cache is faster when you transfer large files, like Blu-Ray ISOs, from another high-speed storage device. Writing data directly to the TLC flash increases flash wear, so WD struck a balance between the two techniques.

SSD caching policies can be a double-edged sword. If data remains in the cache for an extended period of time it increases the odds of a "cache hit" when the system requests data. Fast cache flushing reduces the chances of a cache hit, but it allows newer data to be available in the high-speed SLC. Both methods have advantages, but most enthusiast-focused SSDs use an aggressive flushing policy like the new Black SSD.

WD's controller supports advanced low-density parity check (LDPC) error correction code. In the event of data corruption or loss, the controller tries several less-aggressive soft decode recovery methods that have less of a performance overhead. If that doesn't work, it drops the hammer with a multi-page hardware XOR recovery that results in increased latency. The Black doesn't support hardware-based Full Disk Encryption (FDE) with TCG Opal or Microsoft's eDrive.

Software

The WD Dashboard SSD software package works with the new Black series. The software lets you update the firmware, monitor SMART reporting, and even see performance in real-time. You can also secure erase the SSD if you install it as a secondary storage volume.

Pricing, Warranty, & Endurance


1TB
500GB
250GB
WD Black NVMe SSD
$449.99 - 600 TBW
$229.99 - 300 TBW
$119.99 - 200 TBW
SanDisk Extreme$449.99 - 600 TBW
$229.99 - 300 TBW
N/A

All three capacities carry a 5-year warranty with endurance restrictions that are listed as TBW (terabytes written) in the table above.

Packaging

It is difficult to identify the new NVMe version of the WD Black SSD. Western Digital didn't actually put "NVMe" in the product name, even though the previous version was officially named the "WD Black PCIe SSD." The two products will overlap in the market for a brief time even though WD will discontinue the older PCIe version. The issue isn't unique to WD's flash-based products. The WD Black HDD line has shipped with a hundred (we estimate) different versions over the decade its been on the market.

A Closer Look

All three capacities conform to the M.2 2280 S3 single-sided form factor. The Black NVMe has a regular paper label, but the SanDisk Extreme has a paper-thin metallic label that may (or may not) reduce thermal throttling.

WD placed the controller in the middle of the PCB to aid cooling and reduce the trace length to the flash. This does increase the trace length slightly from the controller to the port, but it reduces the length of the traces connected to the flash.

MORE: Best SSDs

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
18 comments
Comment from the forums
    Your comment
  • Snipergod87
    What does "The 960 EVO uses a five-core ARM design for the 960 Pro and various enterprise variants." mean on page 1
  • AgentLozen
    SniperGod87 said:

    What does "The 960 EVO uses a five-core ARM design for the 960 Pro and various enterprise variants." mean on page 1


    It sounds like the controller that powers both the 960 EVO and the 960 Pro is built on a five-core ARM architecture. Also, Samsung has enterprise level solid state drives built on the same controller.

    edit: In retrospect, I'm not sure if I interpreted your question correctly.
  • photonboy
    "The 960 EVO uses a five-core ARM design for the 960 Pro and various enterprise variants."

    I guess that's been changed to this:
    "WD chose a 28nm tri-core design to target the same market Samsung addresses with the 960 EVO. Samsung uses a five-core ARM design for the 960 Pro and various enterprise variants."
  • Ninjawithagun
    Your benchmarks are wrong. I own a Samsung EVO 960 and am using the latest firmware as well as the most current Samsung NVMe driver for Windows 10. My EVO 960 easily beats the WD Black NVMe in every single benchmark you ran. Something tells me that you did not update your test hardware properly.
  • CRamseyer
    The 960 EVO was updated and retested on December 9th, 2017 for the new burst test results. The results are accurate for the tests. If you use CrystalDisk Mark then your test parameters are different than mine.
  • 2Be_or_Not2Be
    Can it be explained how the Toshiba was #1 in the 70% Mixed Sequential Workload when it finished mid to low in every other test, including the sequential read/write tests?
  • silver12
    In other SSDs reviews you showed the HD Tune Pro test for sequential write. I find that information very interesting. Is there a chance you can add that test?
    Thank you very much
  • mischon123
    Good review. So its basically a broken design with flawed firmware that cannot outperform a 6 year old Samsung product and WD cannot fix the flaws themselves and it took an external firm to tame some of the bugs in the sd design. Now your supposed to pay the same price as you would for a fully working Samsung SSD? Yeahhh, riiiighhht.
  • CRamseyer
    The cache on the 1TB drive is right around 20GB with 128KB blocks. In the future we will publish that data in a line chart. I'm just working on getting a few products retested.
  • Geef
    Does WD use write cache like the Samsung EVO does? Once the cache is used the performance drops off. The Samsung PRO doesn't use cache, its just faster overall. Its a big reason the price difference between the two.
  • CRamseyer
    Yes. The WD/SanDisk cache is called nCache and this product uses the new 3rd generation version.
  • HERETIC-1
    Was often suspicious that some of Toshiba's controllers were Phison-in disguise.
    Hopefully that's not the case here....................................
  • Istarion
    I think this is one of the most pesimistic reviews I've ever read in tomshardware.

    WD did a poor job with its previous "black ssd", with bad performance using a 3rd party controller and setting a high price tag and using the same label as their previous products as it was "fast" when it wasn't, I totally agree with that and was extremely disappointed too, but this is not that product...

    In this case we have a product that:
    -has 960evo-like performance (a bit better overall except for some odd results for some tests)
    -matches 960evo price
    -it's the FIRST CONTROLLER they have done which equates the performance of one of the best considered products in the market (samsung) that has been around wearing the performance crown FOR YEARS
    -it's the first review and it's not even in the market, so these "odd results" may (or may not ofc!) get better when there's availability. In that case, it will outperform samsung 960evo.

    For me the conclusion points are not objective and thus wrong:
    -Some performance anomalies
    -Expensive compared to emerging products
    -Could use some additional firmware tuning

    Whats the difference between point 1 and 3?? "Some performance anomalies" vs "Could use some additional firmware tuning", isn't that the same? If you fine-tune the firmware, you'll fix "odd results", which is the most critical issue (unless there's some other bug). And if you say that the software is "good", this confirms the previous statement I've just said.

    But my favorite is the 2nd one "Expensive compared to emerging products". So you are saying incoming products will have better price. Seriously... You can add that to ALL the reviews!
    I'd say "odd timing" maybe, but for me now this drive is a real alternative to samsung 960evo, at least for today. Ofc next products will outperform the drive! But that's so obvious it doesn't make sense to be part of a conclusion. It's like saying "wait for the incoming products because they will be better"...


    @MISCHON123 I'm sorry but you're quite wrong, I think due to reading only some bits and pieces of the review:
    1. This is a good design with great performance, which has some bugs in certain cases. I can agree to an extent to say "flawed firmware".
    2. Samsung evo 960 is from the end of 2016 (it was announced around september and available around dec / jan of 2017), that is 1 and a half years, not 6.
    3. I don't even understand what you mean with "WD cannot fix the flaws themselves and it took an external firm to tame some of the bugs in the sd design". It makes no sense. Its not what is happening here. You'd better reread the review, you are just wrong :\
    4. I guess that a drive that has similar performance as another drive can have the same price. It has some up and downsides, and the downsides can outwin the upsides in your scenario, but it can be the other way around. I still understand they have some things to address so I could understand asking for a bit less than 960evo, but are we willing to pay more if they fix it? Because if they do, the peformance will be just better.

    I'm extremely happy for WD to take seriously the SSD market and do that inhouse controller. This means more competition and better prices for everyone! Moreover, this is one of the good-old harddrive makers, with good reputation and, from my point of view, respectable engineers.
    Maybe this comes at an odd time, I just hope they keep walking this road, and improve the following controllers. And obviously adjust the pricing when other products appear.

    I can't stop thinking that their first controller just matched crown products performance, thats impressive!!!
  • CRamseyer
    When we say "software" we're talking about the Dashboard software and not the physical SSD.

    "Emerging" in this case was less than 24 hours after the review went online. We knew the other product would hit Newegg/Amazon within days of our writing the review. We didn't know if it would be available before this review went online. It's not quite "wait for the incoming products because they will be better".

    I agree with your closing remarks. In the long term the new architecture will increase competition, impact pricing and be a viable alternative to Samsung. All of those are very good things. WD/SD having a controller is good for performance as well.

    Does that mean we have to recommend the Black 3D NVMe right now? The EX920 is available now, the SX8200 will be next week (reviewers have samples now). We already have an idea of what Samsung's next generation performance looks like from our review of the PM981. WD/SD targeted the 960 EVO in performance and price and that would have been fine for January 2018. If this was a CES launch then I would be cheering, leading the parade with a baton even. This is April and the 960 EVO has lost it's performance crown to a product (and many others very soon) that sell (and will sell) for quite a bit less.
  • Istarion
    Well that's true :) They're 1 generation late, they'll have to adjust pricing. To be fair, we only know the 1tb performance drive, right?
    Let's hope they keep up the pace so we have more competition :3
  • HERETIC-1
    @ISTARION-Personally I think Chris was quite generous with this drive.
    One of the most important measurements I look at is-QD1 4K Read-
    Samsung has been hitting 10K IOPS on it's SATA drives for years-
    I consider that POOR for NVMe.
    Next look at PCMark 8 total bandwidth(what we actually use our drives for)
    POOR results there-when the Intel 760 replaces the 600 the WD ends on the
    bottom.
    To top it all off-where I live they try to sell WD drives dearer than Samsung
    when they have 80 to 90% of the performance..........................
  • Lutfij
    CRAMSEYER,
    Nice article, keep up the good work. This should come to use with my friends.
  • coolitic
    Similar to the 960 EVO, but this has a 5-year warranty whereas the 960 EVO has a 3-year warranty.