WD SN550/SN750 has a NAND switcheroo too!

[CAT-THE-FIFTH]

SN550:
https://www.tomshardware.com/uk/news...f-slc-runs-out
https://www.reddit.com/r/hardware/co...ing_out_sn550/



It seems the SN750 might be having it too:
https://www.reddit.com/r/DataHoarder...b_drives_have/

Seems to affect drives produced since July. Doesn't appear to be a change to QLC,but different TLC NAND.

[Terbinator]

https://www.reddit.com/r/DataHoarder...b_drives_have/

What on Earth is that Sub-Reddit

[watercooled]

Well spotted. Very grim if WD are doing this too. Especially considering how well-liked the SN550 has been.

Edit: Hopefully it's not a widespread thing. From those links there's a possibility it could be an isolated incident or a faulty drive?

For what it's worth, I've just checked a 1TB WD Blue SN550 drive and its NAND part number doesn't match any of those shown, so that's not a reliable way of identifying the problem it seems.

[CAT-THE-FIFTH]

What on Earth is that Sub-Reddit

They look for deals on storage,but at the same time one of the best places to look for specification changes.

Well spotted. Very grim if WD are doing this too. Especially considering how well-liked the SN550 has been.

Edit: Hopefully it's not a widespread thing. From those links there's a possibility it could be an isolated incident or a faulty drive?

For what it's worth, I've just checked a 1TB WD Blue SN550 drive and its NAND part number doesn't match any of those shown, so that's not a reliable way of identifying the problem it seems.

If you look at the links for the SN750,a few people have been affected. My SN700 which is almost the same drive,does not go down to 1000MB/S during long writes!

[watercooled]

I'd like to think it could just be a firmware bug or similar at this point. From the posts it sounds like broken SLC caching. Perhaps even something to do with hardware configuration - it's hard to say from anecdotes, but hopefully we see some proper tests soon.

Edit: Someone with suspect model number drive but expected write performance. https://www.reddit.com/r/DataHoarder...eb2x&context=3

And someone with the old model number but write speed problems. https://www.reddit.com/r/DataHoarder...eb2x&context=3

It would be interesting to know more about the systems being used by those doing the tests. Or a reviewer getting hold of some drives to test it out properly.

[watercooled]

Just been reading a bit more, NewMaxx (a well-known figure when it comes to SSDs if you're not aware) doesn't think NAND is likely to blame. NAND swaps leading to lower write speed is usually down to switching to higher density NAND reducing (often halving) the available planes with a comparable impact on performance, sometimes offset by higher speed per plane with newer NAND. This is likely not the case here as there is apparently no (known) NAND that meets that specification. Newer NAND should increase speed if anything. So perhaps a controller/firmware issue to blame? Hopefully fixable with a patch, if so.

https://www.reddit.com/r/NewMaxx/com...eb2x&context=3 (link doesn't seem to link to the correct post, click to expand the thread)

[Saracen999]

.... Or a reviewer getting hold of some drives to test it out properly.

To be honest, that is what's needed.

By 'reviewer' I don't necessarily mean the conventional definition (as in some journalist writing reviews, though it could be that, too) but a 3rd-party tester that is willing and able to apply a thorough approach to testng, and has sufficient resources to do it properly. Also, preferably, with the credibilty to make the results trustworthy, and the media access to get them out there.

That, for instance, is why I was quite interested in how Steve at GamersNexus went about that PSU investigation, and especally, about what conclusions he was, and wasn't, prepared to draw from what he got. As a 'reviewer' myself, for some 25 years, that is a project I would hesitate to take on, and probably decline to do. It is hard enough doing a normal review, to be able to see the issues through both ends of the telescope in order to be sure any extrapolations you make from what is inevitably limited time and testing and conclusions you draw, are justified from the evidence you have. The trickiest bit is, having got close-up and your nose well and truly stuck into the product in question, to step back far enough to also see it objectively (the other end of the telescope. You kind of need to see things subjectively, but then also to step back and see them objectively. An awful lot of 'reviewers' either don't, or can't.

[watercooled]

That doesn't appear to have stopped swathes of 'journalists' jumping on the reports this morning and pointing the finger at a NAND swap, even though there isn't really sufficient evidence to suggest that at this stage.

Don't get me wrong, it's good it's getting attention which will hopefully pressure WD (or reviewers) into providing some answers, but it's frustrating that so many sites will simply echo what are effectively anecdotes or dubious hypotheses as 'news'.

From what I have seen, all we know at the moment is that some drives, in some configurations, are producing significantly lower than expected performance. One data point is not enough to claim this is related to a part change; as I said, I own a drive with a part number that's different again! There is also some evidence against this, as the hypothesised NAND with fewer available planes doesn't appear to exist.

Given this is a custom WD controller, I reckon it's unlikely WD have significantly changed it for the worse, hardware-wise. Which IMO hints this could be a firmware-related issue, either a configuration change or a bug. Changing from direct-to-TLC to a folding configuration could produce results like this AIUI. And that's assuming this doesn't just turn out to be some system configuration issue as we have seen before.

[kalniel]

That's a fair point. My 970 evo+ lost between 1/3 and 1/2 of it's performance (in benchmarks) between two test points a year apart - on the same drive. Given it's not full, and hasn't had that many writes, I suspect a CPU or even Windows security patch somewhere is to blame in my case - there have been many, including BIOS updates etc. That's not so say switcheroos aren't happening, but it's remarkably hard to compare the software landscape at review time to the environment where you end up using a drive.

[DanceswithUnix]

The thing I find interesting here is that the SN350 WD Green already exists, as an SSD with QLC flash. If this is an official change from WD then it sort of doesn't make sense unless WD felt they had a much better QLC setup that could compete with TLC.

I did wonder when I first saw this if what we were really looking at was some pirate re-labeling of a SN350 so sell it for SN550 pricing. If they get some WD Blue drives and re-package them as Black drives, then the Green ones could even go back out in official WD Blue packaging.

[watercooled]

That's a fair point. My 970 evo+ lost between 1/3 and 1/2 of it's performance (in benchmarks) between two test points a year apart - on the same drive. Given it's not full, and hasn't had that many writes, I suspect a CPU or even Windows security patch somewhere is to blame in my case - there have been many, including BIOS updates etc. That's not so say switcheroos aren't happening, but it's remarkably hard to compare the software landscape at review time to the environment where you end up using a drive.

It's something Windows should do routinely anyway, but try 'optimise drive' on the optimise/defrag utility. Failing that, a secure erase sometimes recovers performance anomalies, but is obviously more of a hassle with having to back-up/restore data etc.

[CAT-THE-FIFTH]

Testing out my 90% full SN700 500GB and with the 1GB CrystalDiskMask test getting around 2500MB/S writes and with the 32GB test nearly 1900MB/S writes. The drive is nearly 3 years old now.

[kalniel]

It's something Windows should do routinely anyway, but try 'optimise drive' on the optimise/defrag utility. Failing that, a secure erase sometimes recovers performance anomalies, but is obviously more of a hassle with having to back-up/restore data etc.

Yeah it's trimmed routinely, and I occasionally use the optimise option either from win defrag or Samsung Magician manually. Will bite the secure erase option one day - day to day performance isn't noticeably slower (I didn't notice when it regressed and CrystalDisk doesn't show the regression, only Magician's IOPS benchmark did).

[CAT-THE-FIFTH]

The SN700/SN750 actually had some of the highest steady state write performance of any PCI-E 3.0 NVME SSD in 500GB/1TB capacities,bar the Samsung Pro series.



What I find impressive,is how little performance has degraded with the SN700 despite being almost 90% full,and coming onto 3 years old. It was around 3000MB/S when new IIRC,when the drive was empty for the 1GB CrystalDiskMark test. Its why I am concerned if there is some problem,as I want to eventually get hold of a 1TB/2TB SN750 or SN850 at some point!

Its why I won't buy QLC drives - reviews from some of the Polish websites with very filled drives,have shown QLC drives to suffer worse than TLC drives as they get filled.



If you read some of the stuff on Reddit people had compared older SN750 drives with newer ones,hence why people are suspecting something is not right:
https://www.reddit.com/r/DataHoarder...b_drives_have/

bout 6 months ago, I purchased a 1TB WD_BLACK SN750 NVMe drive, to use in my ACASIS TB3/USB external enclosure. Everything worked great, and Read/Write speeds on both M1 Mac and Lenovo PC w/Thunderbolt 3 hovered around 2,600MB/2,700MB (Read/Write). This drive does so well in this particular TB3 enclosure, the enclosure manufacturer started recommending it specifically, and I bought two of the ACASIS units. My original drive worked identically in both enclosures.

Today, I decided to pick up a second 1TB SN750 from Best Buy. Ran a quick Blackmagic Disk Speed Test and saw something wrong with Write speed. It struggled to get above 900MB/s. Read did just fine at 2,700MB/s. I figured maybe something was wrong with the enclosure, or even my TB3 cable. Fortunately, I have two identical enclosures and at least a half dozen TB3 cables. Swapping things didn't make any difference.

So, obviously I try on a different computer, this time Windows. Sure enough, Crystal Disk Mark shows degraded Write performance on the "new" drive. CDM always reports about 10% faster than Blackmagic in my experience, but no difference in the overall story. The new drive was only Writing ~1,000MB/s, Read of course did fine at 2,800MB/s. I also tried the WD utility to check/update firmware. Both drives were running the latest 0208 firmware, no update available.

At this point, I had done all possible troubleshooting, steps so I exchanged it for another identical unit at Best Buy. Sad trombone... it has the exact same (poor) Write performance!

So I dunno what to do at this point. The only visible difference between the old/new drive(s) is what appears to be the Julian dating as the initial part of the serial number. The new drives are all "21xx", and my old one is "20xx". Is this particular WD drive a victim of a revised controller and/or other component decontenting? Sure appears that way.

That person had an older SN750,and compared it to two new SN750 which had different production dates,both in an enclosure and in a PC.

Hence you can rule out Windows updates,an apparently the firmware was updated to the same version on both.Since the SN750 is a tweaked,slightly faster SN700 with the same controller/NAND,the fact that writes have gone down to 1000MB/S is really weird. My SN700 on its latest firmwate,latest Windows updates and almost empty is still significantly faster than some of these brand new and empty SN750 drives(32GB test was just under 2000MB/S).

[[GSV]Trig]

WD at it again, someone somewhere needs to lose a job over this sort of thing.

[CAT-THE-FIFTH]

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?