Silent changes to Western Digital’s budget SSD may lower speeds by up to 50%

https://arstechnica.com/gadgets/2021/08/silent-changes-to-western-digitals-budget-ssd-may-lower-speeds-by-up-to-50/

by elsewhen • 5 years ago

386 104 5 years ago

Const-me2• 5 years ago

We only have 4 manufacturers of SSD drives who also making their own chips: Intel, Micron, Samsung, and SK Hynix. Here’s the table: https://en.wikipedia.org/wiki/List_of_solid-state_drive_manu...

I tend to buy Samsung and Intel, both working well in my devices.

kogepathic• 5 years ago

I thought Western Digital had vertically integrated their SSDs after the purchase of SanDisk?

According to that Wikipedia list, Western Digital does manufacture their own flash, through a joint venture with Kioxia.

andruby• 5 years ago

Also, Western Digital is looking at completely buying that Kioxia

https://www.reuters.com/technology/western-digital-advanced-...

justinclift3• 5 years ago

I tend to buy Samsung and Intel, both working well in my devices.

It's a generalisation, but be careful of using new Samsung consumer SATA drives (860, 870) if you're using Linux.

The Linux kernel has workarounds in place for the older ones (840, 850 series). But the workarounds for the newer models are only starting to be looked at this week:

https://bugzilla.kernel.org/show_bug.cgi?id=203475

Without those workarounds (eg disabling NCQ trim), there are data loss problems.

wtallis4• 5 years ago

disabling NCQ trim

Fixed support for queued trim was part of the marketing claims for the 860 series. Whatever's going on with the 860 and newer drives requires deeper investigation, because there's plenty of reason to believe that simply applying the old workaround is not accurately targeting the real problem.

alias_neo2• 5 years ago

Thanks, I had no idea about this. I've been using Samsung drives in all of my machines and servers and I run Linux exclusively. With a mixture of 850, 860 and 870's I must be at least partially affected.

Const-me2• 5 years ago

Thanks for the warning. Fortunately, I’m unaffected because running Windows.

I’m running Linux on top in WSL1 and VMWare.

Also running Linux on ARM devices which have neither SATA nor PCIx. When I need fast disks in them to compile my C++ code, I simply use a network-mounted drive to build, gigabit LAN is good.

aidenn02• 5 years ago

Are 860s new? They are what's in my system and while I don't remember when I upgraded it was definitely pre-COVID...

[edit]

smart says they have 631 days of power-on time, so maybe 2019?

[edit]

I seem to remember having issues a while back but I haven't seen any since I upgraded my firmware to RVT04B6Q

scns3• 5 years ago

Sadly my goto Crucial started doing it too, but they paired it with a huge price drop. Was the P2 NVME IIRC. ~70€ for 1TB.

scns3• 5 years ago

Just read that Samsung changed the BOM on 970 evo plus. Switched to the controller of the 980 pro. Test reults are mixed. 4K QD1 is up ~50->82MB/s, serial writes after exhausting SLC cache (115GB) are nearly halved 1500->800MB/s.

wtallis4• 5 years ago

That's not what happens when you switch to a newer, faster controller. Those performance changes are what happens when you switch to newer NAND flash that's manufactured with twice the per-die capacity, so you have half as many dies for a given drive capacity. (Assuming the newer NAND isn't split into twice as many planes per die to compensate.)

Edit: Looking into this further, TechPowerUp, Computerbase and others have mistaken K90UGY8J5B for K9DUGY8J5B. It's an easy mistake to make, and I've done exactly this before. But it completely explains the performance difference. The D that changed to a 0 signifies a switch from 16 dies per package down to 8 dies per package. The digits signifying the capacity of the package have stayed the same. The "B" at the end signifying the generation has also stayed the same, but Samsung didn't introduce 512Gbit TLC dies until a generation after they introduced 256Gbit TLC dies, so on the smaller dies the "B" means 92L and on the larger dies the "B" means 128L.

randerson• 5 years ago

And Intel just sold their SSD business to SK Hynix, so soon there will only be 3!

fomine32• 5 years ago

This issue happened on WD (acquired SanDisk). So the "just buy NAND manufacturer's SSD" strategy don't work. Also you miss Kioxia!

userbinator• 5 years ago

Products that don't behave the same should really have different part numbers. In this example, perhaps they could call them SN550 and SN550A (B, C, etc.) or even SN551, SN552, ... if the majority of the design hasn't changed.

mirker• 5 years ago

Yes, a great current example is consumer motherboards with Intel’s I225V network controller. The controller is unable to run at 2.5Gbps without a fix. The controller was fixed in revision 3, but a huge number of boards are shipping with revision 2. This is two years after the issue was identified, and there is no way to see which one you will get. Even boards manufactured this year have the old NIC, so I assume vendors are being shady to liquidate old NIC stock.

So you get sold a motherboard with a 2.5X NIC slowdown compared to advertised (it gets downgraded to 1Gbps). Update the figure on the box if you’re going to do that!

sennight• 5 years ago

The most recent example of a "soft recall" would be Gigabyte's exploding PSUs. It makes you wonder how much retail outlets knew, because it is clear that Gigabyte was fully aware of the issue - and silently fixed their design.

xyzzy21• 5 years ago

This happens with products all the time. Strictly "shrink-flation" of consumer goods is the same thing.

One industrial vendor we once partnered with took a product P that had speed S and accuracy A, discontinued that product and replaced with product P1 which was speed S/10 and accuracy A paired with product P2 which was speed S and accuracy A/10. The original P products were precious if you could find them.

belorn• 5 years ago

Silent change in this concept means there is no way to identify P2 from P, and both are sold under the same product name, product code, product specifications, and same product reviews. The later is the key aspect as selling products with the intent to trick the consumer is a defining aspect of fraud when done intentionally for profit purpose.

While shrinkflation often also intend to trick the consumer, the details are usually defined. The packet weight or volume number is changed, the weight cost is change (a list requirement for many products where I live), the number of items has changed. With silent changes however there is no such information available and the consumer just will have to hope that their P is not P2.

ta988• 5 years ago

I'm super happy with my SN850 except those pesky leds that you cant turn off from Linux.

MR4D• 5 years ago

Why is this not fraud?

If you as a company source a device for reviewers and then change the specs in a way that you do not communicate, that is intent to defraud.

If you bought a corvette because all the car reviews said it went 0-60 in 5 seconds, and then you buy one and it takes 10 seconds to do the same thing, you’d have a pretty good argument - especially if everyone else experienced your same situation.

I’d be surprised if a class action suit or FTC investigation doesn’t happens within the next year.

tyingq• 5 years ago

They seem emboldened by the event where they silently introduced SMR into NAS drives. That is, they didn't suffer any real consequences there, so now it's a free-for-all.

alias_neo2• 5 years ago

They must have lost a fair amount of sales?

I previously would never have considered buying Seagate, but since the debacle, when it came to replacing a failed drive in a ZFS array of WD Red CMR drives, I couldn't trust I'd get a drive that would actually resilver so I just replaced them all with Seagate IronWolf NAS drives.

The ops team at my work also avoids WD now, and I no longer buy WD SSDs or external drives because of this distrust.

zozbot234• 5 years ago

One might argue that the changes are marginal and most users won't be affected by them. This change hurts write throughput for large write workloads, but these workloads are rare anyway.

dragontamer• 5 years ago

If you change the specs, people expect your SKU to change. Don't call it the SN550 anymore. Call it the SN551.

There's a whole slew of internet benchmarks on the SN550 and how it performs on a wide variety of benchmarks. For WD to muddle their branding by calling this new, slower drive ALSO an SN550, weakens the brand and causes distrust in builders.

leetcrew• 5 years ago

I don't think there's any requirement that manufacturers guarantee results of third-party reviews for the entire run of a product. it probably just has to meet the specs published by the manufacturer. as long as you still hit that burst read number with some insane queue depth, everything's cool.

the corvette example doesn't really make sense because the manufacturer provides their own 0-60 number. if no one can hit the manufacturer's number in ideal conditions, that's a problem. now if the manufacturer updates their official specs to 0-60 in 10 seconds, that would be pretty weird, but not illegal.

I remember a similar situation with RAM a few years ago. there was some highly sought-after die that could be overclocked way outside the official specs. you could buy a couple different examples of the same SKU and they would have wildly different overclocking limits based on the underlying die. eventually that die wasn't produced anymore, so no one got the insane overclocks. people were really unhappy about that, but at the end of the day, they were relying on behavior that was outside the promised specification.

LeifCarrotson2• 5 years ago

I think it's at least in part because everyone understands (or thinks they understand) 0-60 times for cars.

But when you start talking about multi-level caches single-level cell memory and the shingled magnetic recording debacle and get down to 02031 1T00 SanDisk Flash vs 60523 1T00 SanDisk Flash chips and continuous writing performance of more than 12 GB, eyes begin to glaze over. For most consumers, it's a black box, the implementation details are unknowable and irrelevant.

People would be ticked off if the Corvette had an undersized radiator that caused it to go into limp-home mode any time you took it to the track, we're starting into slightly esoteric "car guy" details. But enough people know what a radiator is and care about cooling system performance that car manufacturers listen. Very few consumers understand the workings of an SSD; manufacturers are under a lot less pressure as a result.

theli0nheart• 5 years ago

You don’t need to understand the workings of a car to get that 0-60 in 10s is worse than 0-60 in 5s.

Likewise, you don’t need to understand the workings of an SSD to get that 10s to copy 1GB of data is worse than 5s to copy 1GB of data.

Implementation details are irrelevant here.

bachmeier• 5 years ago

eyes begin to glaze over

This argument is not especially helpful as a defense against a claim of fraud. It does nothing to explain why you'd give reviewers devices with different performance than the devices you sell.

goodpoint• 5 years ago

But when you start talking about multi-level caches...

No, this is not the point.

The average buyer knows very little about cars or buildings but is protected by regulation.

Consumers should not need to be expert to be protected.

havockatie• 5 years ago

I hate to break it to you but this is EXACTLY what happens with a C7 generation Corvette. The C7 generation base Stingray, GrandSport and Z06 models do well enough to wow in magazine and vlog reviews but when owners got their cars and started taking them to track days they realized they started over heating after not many laps in under 30 mins.

They are obviously performance oriented machines but mostly designed for every day use on the street. They may have big brakes, great suspension and a pretty great engine but the cooling system included is not equipped to handle a track for more than 30 minutes. Within the first month of deliveries people were freaking out. After 30 mins the car would go into limp mode and you were lucky if you got on the track at all again that day.

But guess what? The reviewers never experienced this.

antasvara• 5 years ago

The real question is what Western Digital says the speed for the drive is. I would guess that the changes don't touch the stated top speed, as the article seems to indicate that this only slows down write speeds after the cache is filled.

If Western Digital specified an average write speed, this might be affected. However, that average could technically not have changed depending on how they determined this average speed (for example, they could have determined this average by downloading and uploading smaller files).

I'd say that Western Digital isn't affecting the 0-60 time in ideal conditions, just decreasing the ability of the drive to attain that 0-60 time in less than ideal conditions.

Dylan168072• 5 years ago

The real question is what Western Digital says the speed for the drive is.

They say "Up to 1,800 MB/s"

So if they decided to put in flash similar to a samsung 860 QVO, which drops down to 80 MB/s when the cache is full, that would still satisfy the spec.

webmobdev• 5 years ago

You are right - it should be treated as fraud. It's a common business practice nowadays to bait and switch - first release good quality products, get good reviews for it and then use those good reviews to market your product. And after it has sold a certain quantity and established a brand value, start using cheaper and inferior quality components and extract more profit.

sneak• 5 years ago

It is fraud, but laws against fraud are not generally enforced against large businesses in important industrial supply chains unless they consistently and repeatedly flaunt it.

I'd be surprised if there is a class action or FTC penalty.

AtlasBarfed• 5 years ago

There's a very good reason Consumer Reports buys from normal retail channels.

No Consumer Reports like publication would survive for this stuff. They'd be so far behind the other reviewers, even if they had better quality data.

cableshaft2• 5 years ago

This happens to video games also. Publishers give review copies with microtransactions not included, the reviewers review it, and then shortly after release they add their free-to-play garbage so they get reviews that don't take the microtransactions into account and then they add it later.

The also do it during ESRB submission to get around having to have a label that mentions in-game purchases. I'm surprised ESRB lets them get away with it, actually.

Jimquisition on it: https://youtu.be/9LeZ89_u2Gc

Article on it: https://www.tweaktown.com/news/66889/publishers-delaying-mic...

pulse7• 5 years ago

What are those microtransactions doing? Crypto mining?

michaelt• 5 years ago

The reviewers get sent an account with everything enabled, and have a great time playing as Darth Vader Or Han Solo Or Whatever.

The general public buy the game, and find if you want to play as Darth Vader or Han Solo unlocking them costs $20 each.

dharmab• 5 years ago

"Microtransaction" means small purchases. Such as buying a randomized pack of in-game unlocks for $1 each.

https://en.wikipedia.org/wiki/Microtransaction

some_random• 5 years ago

ESRB is industry controlled iirc, and microtransations/lootboxes are a huge cash flow for the largest games companies

wiz21c• 5 years ago

Et voilà. Conflict of interests all the way...

plorkyeran• 5 years ago

Yeah, it's voluntary self-regulation that was establish to stave off laws being passed in the early 90s during the panic over Mortal Kombat, Doom, etc. Presumably if legislators start looking into microtransactions and rating tricks then the ESRB will start taking them much more seriously.

opdahl• 5 years ago

One of the latest events like this is Capcoms Resident Evil Village which launched with a DRM that added terrible stuttering to the PC version.

They only admitted to this and promised a fix when a cracked version came out that had the DRM removed and it didn’t have any stuttering [1].

[1] https://www.gamesradar.com/resident-evil-village-pc-patch-la...

smoldesu• 5 years ago

The folks at Digital Foundry did a phenomenal job breaking down the specifics, as usual. I strongly suggest checking out their video if you're interested in the stuff:

https://youtu.be/UXZGCwAJpbM

gentleman11• 5 years ago

Capcoms pc ports have always been… I remember my dmc3 that didn’t work at all (maybe a fan patch helped?), and the frame pacing in re4 on pc giving my wife and I migraines… or dmc5 deleting my save file… or re2 remake requiring 2 year old drivers to avoid crashes… I have friends who have issues with the street fighters on pc as well. That’s just off the top of my head.

jbverschoor• 5 years ago

Sounds like the diesel scandal

dijit• 5 years ago

I work on an upcoming free to play game, we will have microtransactions on cosmetics and possibly new character archetypes.

Microtransactions are not in the early builds of the game because we literally hadn't made the content for it yet, we were focusing on the game mechanics.

The store is a lot of work in a game, and the cosmetics themselves are mostly content work which is not hyper critical to have in place (unlike, say, matchmaking or gun mechanics).

I'm not saying it hasn't been done nefariously; just that: The game I'm making doesn't have microtransactions yet because we worked on other stuff.

But we're not lying about them being there, they have to be there otherwise we'll go bankrupt paying the server costs and failing to recoup R&D costs.

Dayshine• 5 years ago

You should be ensuring that all reviewers mention that this is an early access dev build with all paid content unlocked then, right?

I don't think people are talking about cosmetic micro-transaction though.

cableshaft2• 5 years ago

I'm not saying all microtransactions are bad necessarily. I've also worked on a couple of games that have had them, although we did have them on launch (and added more to it later).

But it is kind of a shitty move if you let reviewers review your game and let them believe there won't be microtransactions and then you add them later.

Now if your game is a free game to begin with, it's sort of expected you'll add something in there to make money at some point. But most of these big AAA games are $60-100+ to buy the game to begin with, so there's not an easy assumption to make. Also they almost always have it available a few weeks after launch, as if they were just waiting to get past ESRB and reviewers without mentioning those things.

Finally, I've worked on console games and handled ESRB submissions for our games. You're not supposed to leave anything out. We got dinged once for not mentioning that the card art on one of our cards had a character smoking a pipe on it, and another card where the anime girl had a slightly exposed nipple (we didn't even notice, it was a port of a game by another developer and the card art is busy enough you kind of have to be looking for it... if we had known, we would have told the developer to edit those out and the game wouldn't have had the 'smoking' and 'nudity' label on our rating). The video you submit is supposed to show everything in the game. And yet these companies aren't doing that so they can keep the in-game purchases label off their game.

If you're trying to hide something in your game from certain parties, you're being shitty and you should stop, is the moral of the story. If you're not you're probably fine.

For the ESRB rating for the game I worked on, here's the back of the box for proof: https://www.vgchartz.com/games/boxart/full_6178352AmericaBac...

tmp_anon_22• 5 years ago

Bottom line is reviewers are reviewing one version of the game, and players are getting a sometimes significantly different version. That is anti-consumer behavior no matter how you cut it.

Some responsibility does also fall on reviewers for being aware and where appropriate mentioning this fact.

rrrrrrrrrrrryan• 5 years ago

Phone manufacturers do this as well. Release a phone to reviewers with awesome performance and battery life (by aggressively killing background apps), then a month later, release a build that doesn't do those things so people don't complain about missed notifications, etc.

Hell, even Apple nuked the noise cancelling in the Airpods Pro right after launch [1], possibly for battery reasons or to prevent hearing damage lawsuits, but by all accounts they had jaw-dropping noise cancelling when the reviewers were writing their initial reviews.

[1] https://www.theverge.com/2020/1/17/21069953/apple-airpods-pr...

sebazzz• 5 years ago

I see a market for a site called "The late review".

Abishek_Muthian• 5 years ago

IMO with games it's preplanned fraud for profit, With hardware it's due to pandemic induced chip shortage but I guess it's still a fraud when product states something and delivers another.

I was expecting that the silent feature reductions in the automobiles due to chip-shortage would seep into another hardware products sooner or later, In fact I was trying to validate whether there's a need to track 'What has been removed due to chip shortage'[1] and it seems like there's good reason to do it.

[1] https://needgap.com/problems/234-what-has-been-removed-due-t... (Disclaimer: I created this platform).

aurizon2• 5 years ago

Wanna buy a one legged duck? People who will want this are low end system builders who can add the "Has an SSD" who market to people have no idea how fast a 2 legged duck can run....

bserge• 5 years ago

*fly. So it makes no difference until the duck lands :D

aurizon2• 5 years ago

Maybe the duck has trouble taking off too? In effect they sell a crippled product for less $$ to the cheap system builders who market on price to people impressed by SSD, who know no better

zucker42• 5 years ago

This was a problem with the Adata XPG SX8200 Pro, which LTT covered in this video[1].

[1] https://youtu.be/K07sEM6y4Uc

post_break• 5 years ago

Ship the top tier hardware NVME drives to reviewers, wait a few months then start swapping out slower memory chips, controllers, go through multiple revisions while keeping the same name so when gamers look for reviews they see blinding speeds only to be duped. It's incredible what drive manufacturers are getting away with.

HauntingPin• 5 years ago

I've also seen this happen with laptops. They'll have a decent Samsung NVMe SSD when reviewers get their hands on the devices, but then after release, they switch it out for a cheaper one that performs much worse, particularly in longer workloads.

p_j_w• 5 years ago

It's incredible what drive manufacturers are getting away with.

I think it's more incredible that we continue to elect governments that let them get away with it.

exporectomy2• 5 years ago

This doesn't need more regulation. Just stop trusting professional reviewers who are not bound by the existing false advertising laws that the manufacturers/retails already are. If you want to know how something performs, read the specs published by the manufacturer. If they're too shy to say how slow it is, then there's a reason for that!

vizzier• 5 years ago

Linus Tech Tips recently did a video on this practice[1]. Almost every major component was changed but still sold as the same drive. In summary, performance was different though not necessarily in a major negative way. In this case it was an ADATA NVME drive.

[1] https://www.youtube.com/watch?v=K07sEM6y4Uc

bitL• 5 years ago

I bought a 1TB WD Blue M.2 SSD like the one on the picture and my computer has issues booting from it when all USB devices are plugged in (x299/10980XE). It looks like more people have issues booting from it as well, based on the feedback I saw on the seller site. How can one botch SSD that way?

y04nn• 5 years ago

Unbelievable, but is there a datasheet available and does the datasheet have read/write speed figures?

Edit: [1] The datasheet shows "up to" speeds, which means that it is the disk cache speed and not the NAND speed? Quite disappointing.

[1] https://documents.westerndigital.com/content/dam/doc-library...

deelowe• 5 years ago

Don't expect useful specs to be available for consumer drives. The datasheets have slowly been turned into marketing materials over the past 10 years. Only the devices targeted for business use have useful specs and even then, key details such as MTBF or similar can be missing.

0-_-0• 5 years ago

Quick summary: The drive is 1TB with a 12 GB SLC cache that's capable of 2160 MBps write speeds. When you write more than 12GB in a short time to the drive you can run out of this cache and speeds drop to 390 MBps. The dive was previously tested as capable of 610 MBps in this scenario. (Read speeds are fast in every scenario.)

burnished• 5 years ago

This is kind of a hyperbolic re-write of the linked article at Tom's Hardware[1], which is I guess originally sourced (by both articles) at a Chinese site called expreview. The linked article at Tom's Hardware has more information and more context that is, confusingly, left out of the top level article.

[1] https://www.tomshardware.com/news/wd-blue-sn550-ssd-performa...

p.s is it possible to change what the link is to? it seems sort of pointless to have a discussion on an article that appears to be clickbait, in that it adds nothing substantive and in fact neglects to include a lot of information.

aidenn02• 5 years ago

Note that it's not just SSDs that can have completely different components on the same part. I've seen many USB wifi dongles with the same model number use completely different wifi chips internally.

theropost• 5 years ago

I think I'm done with WD from now on. The whole WD MyBooklive remote factory result fiasco annoyed me as well - for those impacted they offered a 40% discount on one of their new MyBookLive devices, as long as you send the old drive back (shipping paid by user).

kstrauser• 5 years ago

WD is on my personal blacklist for now. Don't forget last year when they lied and marketed SMR drives as NAS-compatible: https://www.tomshardware.com/news/wd-fesses-up-some-red-hdds...

I can forgive hardware failures. Packing a few trillion bits into a box and making them perfectly reliable is a hard job. What I absolutely cannot and will not forgive is Western Digital lying about their specifications. If I can't trust their words, damned if I'm gonna trust their products.

andrewstuart2• 5 years ago

We ought to have the equivalent of crowdsourced semantic versioning and CVEs for products and reviews. As a consumer, I should be able to know _easily_ if any breaking changes have happened that could render a specific set of reviews or benchmarks obsolete, and just as easily if a specific issue is resolved in future revisions.

Too often I've e.g. bought an Amazon product I've bought before, only to receive Revision 2 that doesn't fit my case any more, but that the company silently rewrote descriptions for just to keep their review score.

I understand why people want to keep their review scores, etc, and to some extent they _should_ keep them, but not at the product level, when the product changes.

cwizou• 5 years ago

This happens a lot with flash in general, mostly because those who manufacture the end user product don't usually make the NAND (or don't always use their own).

I had the issue with "high end" USB sticks a while back during a review. The speeds given are completely irrelevant and always based on best case scenarios. And while a few manufacturers stick to the same formula for controller + NAND for the life of the product (say Sandisk), others just don't care and will silently switch both the controller and NAND with 0 indication on packaging (no rev., no nothing).

On SSDs I can't remember seeing that though, but then again, this is Western Digital we're talking about. A company best known for, let's say, being a very late entrant to the SSD market.

Edit : At least they (kinda) owned up to it after being caught, quoting the updated article : "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,"

BearOso• 5 years ago

Also, a warning on Western Digital”s SN750 SE series. These are PCIe 4.0 protocol ones that claim to outperform the regular SN750 drives. In reality they’re slower because they’re DRAM-less. Last I checked they were the same price, so be careful to not get the SE model.

LinAGKar• 5 years ago

How much of a difference does the DRAM cache make anyway, given that there is already a significantly larger disk cache in the system RAM, which is also closer to the CPU?

fomine32• 5 years ago

On SSD market, silent change in same model and performance degrade is continuously happen. Any maker should change model number.

reviews-chat• 5 years ago

Not fair

Silent changes to Western Digital’s budget SSD may lower speeds by up to 50%

Related Stories