WD SN550/SN750 has a NAND switcheroo too!

[watercooled]

Production dates, countries and part numbers change all the time, that by itself doesn't indicate a specification or performance change. It would be good to get to the bottom of it, but especially in the case of the SN750, I'm hesitant to point the finger at anything sinister as it seems the SLC is outright not working properly. Even a NAND swap would not cause a performance degradation that severe.

WD also state the write speed on their website: https://shop.westerndigital.com/en-g...sd#WDS100T3X0C

3000MB/s for the 1TB version. Not meeting specification suggests something is broken.

[DanceswithUnix]

Testing the Blackmagic Disk test,I was getting between 1200MB/S to 2200MB/S for writes,dependent on the test run,and around 2800MB/S for reads. This is on a 90% full 500GB SN700. Something is definitely not right with some of these empty SN750 1TB drives then!

DWL has a SN750 1TB?? Maybe they can test out their one with the same software and see if any weirdness is happening with their existing firmware?

I do, though that isn't my main machine and I'm a tad busy right now so might be a bit tricky. Is it crystal diskmark you wanted? If so I shall try and get some readings.

[watercooled]

Quick CDM 6.0.2 test on a SN700 500GB, 1GB test size, 1 run:

Sequential read: 3339.5
Sequential write: 2521.3

[CAT-THE-FIFTH]

Production dates, countries and part numbers change all the time, that by itself doesn't indicate a specification or performance change. It would be good to get to the bottom of it, but especially in the case of the SN750, I'm hesitant to point the finger at anything sinister as it seems the SLC is outright not working properly. Even a NAND swap would not cause a performance degradation that severe.

WD also state the write speed on their website: https://shop.westerndigital.com/en-g...sd#WDS100T3X0C

3000MB/s for the 1TB version. Not meeting specification suggests something is broken.

WD can change specifications stealthily - the WD Green controller was changed totally and they newer actually officially admitted it(second generation Green SSDs were much slower because they used a controller which was used in USB Drives,but it also drew much less power).

However,my main concern is the fact the guy who tested the two newer SN750 1TB drives noted they were the same firmware revision as their older one,just a new production revision.

So something seems off with the new ones. My SN700 has the same controller and AFAIK the same NAND too. The SN750 is a tweaked SN700. Using the same tests,on a very full drive,I can hit at least 1200MB/S on a smaller drive which is almost full.Plus that is the other issue,the SN750 1TB steady state is well over 1400MB/S for larger writes.

Going down to 1000MB/S on a brand new drive,means something is going funky....unless WD is doing what Adata did with the SX8200 PRO,and have slightly different specs under the same generic model number? Because when the SX8200 PRO had performance inconsistencies it was because Adata was having different component parts in different batches.

I do, though that isn't my main machine and I'm a tad busy right now so might be a bit tricky. Is it crystal diskmark you wanted? If so I shall try and get some readings.

Quick CDM 6.0.2 test on a SN700 500GB, 1GB test size, 1 run:

Sequential read: 3339.5
Sequential write: 2521.3

That is almost the same as my nearly full SN700 500GB.

This is the Blackmagic Disk test:
https://www.blackmagicdesign.com/sup...e-and-playback

(Desktop Video 12.1). Will be interesting to see if both of you can run the test(5GB files) and see what numbers you get!

[watercooled]

I have a SN550 that has a different part number to any of those listed. Part numbers can be minor hardware revisions, or even down to different production/finishing facilities.

Obviously I'm not saying there's nothing happening, because *something* is wrong. I'm only saying that part number or revision number changes *by themselves* don't mean much, because they happen all the time with electronics. The part number changes could well be coincidental and distracting from the real cause.

As before, a major problem with blaming a NAND change, is that there it seems there is no known NAND that would actually result in those sorts of speed reductions. Switching to higher-capacity NAND with fewer planes happens a lot with certain manufacturers, but WD/Kioxia don't appear to make anything that fits that description.

[CAT-THE-FIFTH]

It's a shame the guy didn't get some pictures of the drives themselves to see if the PCBs have changed in anyway.

[CAT-THE-FIFTH]

WD has confirmed the SN550 has been changed since June:
https://arstechnica.com/gadgets/2021...s-by-up-to-50/

[watercooled]

"For greater transparency going forward, if we make a change to an existing internal SSD, we commit to introducing a new model number whenever any related published specifications are impacted,"

Well that's something I guess. Would be nice if they still offered the existing SN550 that was so well-liked, though.

[CAT-THE-FIFTH]

"For greater transparency going forward, if we make a change to an existing internal SSD, we commit to introducing a new model number whenever any related published specifications are impacted,"

Well that's something I guess. Would be nice if they still offered the existing SN550 that was so well-liked, though.

I am concerned about the SN750 now!

[watercooled]

The SN750 with the suspect model number looks like it has existed alongside the 'proper' one since at least 2019 according to the data sheet: https://documents.westerndigital.com...me-ssd-idk.pdf

I wouldn't worry about it too much, just avoid that model number. Scan, Amazon, and the WD store all list the original model number. The 4TB is also a recent addition to the range, which is part of the original range, not the weird underperforming one.

Edit: For reference, the existing range's model numbers end in 3X0C, the weird range ends in NC-WRSN.

[kompukare]

CB report the Samsung 970 EVO Plus also having had component changes:
https://www.computerbase.de/2021-08/...mponenten-aus/
Some benches remain the same, but other are worse:

Probably wouldn't complain about that. Yes, Seq write is worse but 4K-64Thrd is a good bit better. Overall score is better.

[CAT-THE-FIFTH]

Sequential write is still important:
https://www.extremetech.com/computin...with-slow-ssds

The issue is that its going to bog down movement of large files,and also if you intend to do video work. If you read that link,if you intend to do video stuff it really becomes a big problem.

Edit!!

From the computerbase.de article:

The benchmarks carried out show a diverse picture for the performance comparison of the new with the old 970 Evo Plus. For example, the new version with the Elpis controller achieves better results overall in CrystalDiskMark and in the AS SSD benchmark. The significantly higher rate for the RND4K Q1T1 of just under 82 MB / s is noticeable, while the old version only achieved around 56 MB / s. The AS SSD benchmark also confirms that the new version has a higher 4K performance, but poorer latencies. There are only minor differences in the copy tests of the software, but overall they are in favor of the old version.

In the case of even larger transfers, however, the old version with the faster continuous transfer rate would clearly take the lead. The advantages of the new version are again up to 115 GB.

So that means for larger file transfers its worse. Upto 115GB both are essentially a tie. Above that the write speed drops by nearly half!

Even the "alternate" version of the SN750 does not drop below 1000MB/S!

[watercooled]

The revised 970 Evo Plus uses the Elpis controller from the 980 Pro (but likely restricted to PCIe 3.0) and 512b V6 NAND. Sustained sequential is lower, however initial cached speed is considerably higher, and the cache is larger. Check the HD Tune write graphs: https://www.techpowerup.com/286008/e...0-evo-plus-ssd

I'd describe this one as more of a sideways move, and Samsung do at least appear to have tried to offset performance loss due to the NAND change. Even over a 200GB span, write performance averages about the same as the existing drive.

[CAT-THE-FIFTH]

The 970 series was already worse than the SN750 over sustained writes,in 1TB and 500GB capacities. Computerbase.de say over 115GB things start to shift in favour of the older version,but they also noted latency was a bit worse too.

So as long as the SN750 we get in the UK does not get changed to the one in the US,it probably still works out better for me. But even the US version of the SN750 seems to keep to around 1000MB/S,so it still seems a tad better than the new 970 EVO dropping down to 800MB/S!

What annoys me is that the SN750,970 EVO,etc are designed as performance PCI-E 3.0 SSDs,and people might use them for 4K video work,etc which can generate huge writes(100s of GB). So its really taking the mickey these companies just don't communicate these changes more transparently,as it would force a retest of the SSDs by reviewers.

Its almost like false advertising,ie,seed different models to reviewers and different ones to consumers.

[watercooled]

The weird SN750 isn't a US replacement it seems, it's just available in some stores, and seemingly has been for years.

Things do start to move back in favour of the old 970 Evo Plus after ~115GB because that's where the cache ends, but it takes until around 200GB for the old drive to actually overtake it again, because the cache is so much faster on the new drive. ~1750MB/s on the old drive vs ~2600MB/s on the new drive, likely due to the controller swap and matching the read speed.

I'm not going to say it's conclusively better, but it could be advantageous for some people. Yes, the cache will shrink as the drive fills, but even if the cache shrinks to a third of its size, e.g. on a 2/3 full drive assuming it's allocated that way, in theory it would still match the old cache in empty-drive size and surpass it in speed.

[CAT-THE-FIFTH]

The problem is if the cache runs out at 115GB,it might actually start to have bigger issues with more variable size files. The test only really used a larger video file.

Well in my case I want to move to dual NVME SSDs(and I use them until they break),to move around some of my modded game installs,which are over 200GB. My current Fallout 4,and Vortex mod folders are nearly 300GB. I can see it get worse for Starfield! Since I want to be able to make constant backups when I move to dual NVME SSDs,not sure if I want reduced write performance. Its why I spent extra on a B450 mini-ITX motherboard which could run dual NVME SSDs.

Plus at some point,I might do 4K video stuff,once the Covid stuff gets better and I can make it to some airshows. Its why I went with a Ryzen 7 3700X over a Ryzen 5 5600X! That can generate lots of files(100s of GB),so I need to think ahead if I am buying an SSD. For example I only retired the SSD I won on Hexus in 2012,last year!

Hence the 970 EVO new version doesn't cut the mustard for me,but neither did the original version as the SN700/SN750 seemed better over longer write periods until you went to the 2TB version. I am not going to notice quick burst speed differences,when they are already fast,but I will notice it over much larger writes.

But I might skip both and look at the newer PCI-E 4.0 models,but even that I prefer consistency over burst speeds.