The Western Digital WD Black SN750 SSD Review: Why Fix What Isn't Broken?
by Billy Tallis on January 18, 2019 8:01 AM ESTRandom Read Performance
Our first test of random read performance uses very short bursts of operations issued one at a time with no queuing. The drives are given enough idle time between bursts to yield an overall duty cycle of 20%, so thermal throttling is impossible. Each burst consists of a total of 32MB of 4kB random reads, from a 16GB span of the disk. The total data read is 1GB.
The Toshiba/SanDisk 3D NAND has consistently shown worse burst random read performance than its competition, and the WD Black SN750 does nothing to change that pattern. The SN750 is insignificantly faster than last year's model, but both the Plextor M9Pe and Corsair MP510 get better read latency out of this NAND by pairing it with different controllers.
Our sustained random read performance is similar to the random read test from our 2015 test suite: queue depths from 1 to 32 are tested, and the average performance and power efficiency across QD1, QD2 and QD4 are reported as the primary scores. Each queue depth is tested for one minute or 32GB of data transferred, whichever is shorter. After each queue depth is tested, the drive is given up to one minute to cool off so that the higher queue depths are unlikely to be affected by accumulated heat build-up. The individual read operations are again 4kB, and cover a 64GB span of the drive.
On the longer random read test that also brings in some higher queue depths, the WD Black SN750 is still quite a bit slower than the best combinations of Silicon Motion controller and Micron NAND, but at least this time the drives that use the same Toshiba/SanDisk BiCS3 NAND with different controllers aren't so far ahead of the WD Black.
 |
|||||||||
| Power Efficiency in MB/s/W | Average Power in W | ||||||||
The WD Black SN750 is just a touch more efficient than its predecessor on the random read test, allowing it to retake a small lead among TLC-based drives here rather than being in a tie with the Toshiba XG6.
 |
|||||||||
When the WD Black SN750 is pitted against most of last year's competition, it does at least catch up if not pull ahead in random read performance at higher queue depths, while never requiring as much power. But the Silicon Motion SM2262EN drives are faster at all queue depths, though not in proportion to how much more power they require.
At low queue depths the random read performance of the WD Black SN750 is down in SATA performance territory where some drives can beat it on performance and power consumption. But at higher queue depths, it is obviously the most efficient NAND-based SSD we've tested for random reads.
Random Write Performance
Our test of random write burst performance is structured similarly to the random read burst test, but each burst is only 4MB and the total test length is 128MB. The 4kB random write operations are distributed over a 16GB span of the drive, and the operations are issued one at a time with no queuing.
The WD Black SN750 shows a small regression in burst random write performance compared to last year's model. The Corsair MP510 and Silicon Motion SM2262EN engineering sample are nearly tied for first place, with the WD Blacks about 15% slower.
As with the sustained random read test, our sustained 4kB random write test runs for up to one minute or 32GB per queue depth, covering a 64GB span of the drive and giving the drive up to 1 minute of idle time between queue depths to allow for write caches to be flushed and for the drive to cool down.
On the longer random write test that brings in higher queue depths, the WD Black SN750 is roughly tied with the SM2262EN sample for first place and is a few percent faster than competitors like the Samsung 970 EVO and Corsair MP510.
 |
|||||||||
| Power Efficiency in MB/s/W | Average Power in W | ||||||||
The SN750 is the most power efficient NVMe drive on our random write test, pulling slightly ahead of the Toshiba XG6 and the older WD Black, and maintaining a substantial lead over most other high-end NVMe drives.
 |
|||||||||
The WD Black SN750 hits full speed for random writes with a queue depth of 4 or higher, and maintains steady performance for the rest of the test while drawing just over 3 W.
The WD Black SN750 is very close to the being the most efficient flash-based SSD for random writes that we've ever tested, but there are a few data points in the archive that are slightly faster at similar power levels.
Facebook
Twitter
RSS
54 Comments
View All Comments
joesiv - Friday, January 18, 2019 - link
Micron was the manufacturer I was referring to.Other brands we've used which didn't exhibit the same poor endurance, ADATA, Kingston, Swissbit, Crucial
Some of them probably even use Micron NAND. I bet the NAND is fine on the Micron model we were using, perhaps the hardware is good but the software (firmware) wasn't? Of course we haven't tested every brand/model as our requirements were very specific, so I am sure there are other Micron models that are totally fine (kind of why i'd love to see anandtech include some endurance results, to help weed out the outliers)
sdsdv10 - Friday, January 18, 2019 - link
Interesting you write that Micron has problems and Crucial doesn't, as Crucial is just a consumer brand name for Micron Technology Inc.joesiv - Friday, January 18, 2019 - link
Well they were different models. The crucial was an old model that we were replacing with something new, since the old crucial drives were no longer available. It would be interesting to compare a crucial equivalent model though, I wonder if they share firmware.sovking - Friday, January 18, 2019 - link
Of course, these improvement will be welcomed, and I would like to see more in clear the steady state behaviour too.Regarding the endurance, we should take into account that most of these reviews are about consumer products. An NVME SSD for enterprise market has totally different performance: e.g regular steady state performance, higher endurance, higher reliability and so on. Sometimes, it's possible to find lightly used enterprise NVME drives at bargain price or at the cost of consumer drive: when this happens I prefer these drives.
joesiv - Friday, January 18, 2019 - link
I think the role of a "consumer" is not perfectly defined these days. Are they the same as a "power user?" It would seem that more and more consumers are starting to do more and more serious workloads on their PCs. Obviously this is anecdotal, but with all the processing power at our disposal these days ("consumer" CPU's having 16 threads). People probably don't even know what the applications or services that they are running on their PC are doing.For example, a lot of commonly used applications will be running with a database system as their backend, whether it be a more simple sqlite database, or something more serious, those can be very write heavy, and they're often configured by the application without the user even knowing it. I'll bet that a lot of users even have web services running on their PC's, without actually thinking about it, all these API's that allow you to connect to your mobile devices/streaming appliances.
I'll bet a lot of people reading anandtech reviews even have their PC's running as a fileserver, or have a dedicated machine for such duties.
A lot of this stuff is stuff is stuff would be considered "enterprise" computing of yester-years. Why does anandtech run transcoding, rendering and "destroyer" style tests in their "consumer" reviews? Because it's relevant to some portion of the purchasing community.
Oxford Guy - Friday, January 18, 2019 - link
Considering how consumer parts have had endurance problems...Examples: OCZ Vertex 2 (with 64-bit NAND), Samsung 840 128 (terrible steady state performance, too), Samsung 840 and 840 EVO series (read speed loss), etc.
Endurance isn't just a matter of whether or not the drive dies or it has a lot of cell death. It's also a matter of performance consistency over time.
joesiv - Friday, January 18, 2019 - link
I agree, I have bad memories of the early days of SSD's. I purchased a first generation intel SSD for $1000 (CND), the speeds were tested as being amazing compared to anything else on the market. But given the early learning curves with NAND controllers, and whatever the like, performance was terrible in the real world. I wasn't even able to upgrade the firmware since it was a first generation product, and only the subsequent versions supported the updates.Things have gotten better, but from my experience, it's been a rough road. Some manufacturers are a lot better than others for firmware development, and believe it or not a bug in the firmware can tank performance, or even tank your reliability, since the firmware is what controls wear leveling, and other new fangled features to give the maximum performance.
There are MLC drives that work in SLC mode dynamically to aid in performance, and other drives that are MLC NAND running SLC mode which have a hybrid endurance between the two. Some older drives did driver level compression to reduce NAND writes, while theoretically great, can cause problems for reliability if there are any cases where the data doesn't get committed correctly, especially in poor power conditions. Firmware bugs are rarely talked about, but a firmware bug could cause garbage collection to occur too often, which will take your performance and reliability.
gglaw - Friday, January 18, 2019 - link
With current gen 3D NAND, it would take an incredible amount of writes to test endurance and the regional wholeseller RMA data averaged over hundreds of thousands of SSD's sold is much more representative than AT testing endurance on 1 drive they receive as a sample. It appears most SSD RMA's are NOT from using up the endurance cycles so that would make a 1 sample test by AT even less meaningful. If they happen to get a dud when 99% of that same model has a very good reliability history based on the broader market it would just make thousands of AT readers base their purchasing decision based on a sample size of 1.Billy Tallis - Friday, January 18, 2019 - link
P/E ratings are highly dependent on what kind of error correction the NAND is used with. Even under pressure, the NAND manufacturers won't be able to give us more than just ballpark figures that would be tough to fairly compare between manufacturers.Last year (I think around when the first QLC drives showed up) I started recording SMART data before and after each phase of testing. I haven't written any code to parse and analyze that information yet, but it's on my to-do list.
I don't think the usual consumer SSD test suite does enough total drive writes to move the SMART indicators enough to form meaningful projections about write endurance and drive lifetime. To do that, I would have to set up another system to do long-term endurance testing on several drives at once. That's also on our wishlist, but it's a relatively low priority given the extra equipment and time requirements.
joesiv - Friday, January 18, 2019 - link
@gglaw, @Billy Tallis, you guys are right, it's hard to get firm reliability numbers based off a short, small sample test. But to be honest, its' better than nothing. And as I said, seeing one example of an outlier that performs badly on the bench for the test would validate it's usefulness.gglaw, you are totally right, there is more to reliability than PE Cycles, I gave the examples of a drive that under our testing failed, with a life expectancy under a year, the same test scenario (which was a heavy real world workload for our product) on other similar rated drives did not fail the test. But I didn't mention that we had huge realiability issues with our previous drives (Kingston), where they were no where near the end of their endurance ratings, but were failing for other causes. Kingston attributed a lot of the failures to firmware bugs that weren't traceable in SMART data, and in some cases pure hardware failure.
Billy, yes in general you're right, it's hard to get meaningful projections for a short period of time, this is especially the case if you use percent life used as a metric (1-100). However, it's not too bad if you can get the PE Cycles, which typically are 3000 for MLC, and in some cases 2500 for 3D NAND, instead of waiting months for a single change in percent life change, we have seen drives go through 1 PE Cycle a day, which would give us around 8 years of product life (baring other failures), we were going through 5-6 PE Cycles a day on the Micron drive, which was a huge warning sign. That would be a great case for anandtech finding the poor endurance outliers.