Samsung Is the Latest SSD Manufacturer Caught Cheating Its Customers
In the past 11 days, both Crucial and Western Digital have been caught swapping the TLC NAND used for certain products with inferior QLC NAND without updating product SKUs or informing reviewers that this change was happening. Shipping one product to reviewers and a different product to consumers is unacceptable and we recently recommended that readers buy SSDs from Samsung or Intel in lieu of WD or Crucial.
As of today, we have to take Samsung off that list. One difference in this situation is that Samsung isn’t swapping TLC for QLC — it’s swapping the drive controller + TLC for a different, inferior drive controller and different TLC. The net effect is still a steep performance decline in certain tests. We’ve asked Intel to specifically confirm it does not engage in this kind of consumer-hostile behavior and will report back if it does.
The other beats of this story are familiar. Computerbase.de reports on a YouTube Channel, 潮玩客, which compared two different versions of the Samsung 970 Plus. Both drives are labeled with the same sticker declaring them to be a 970EVO Plus, but the part numbers are different. One drive is labeled the MZVLB1T0HBLR (older, good) and one is the MZVL21T0HBLU (newer, inferior).
Peel the sticker back, and the chips underneath are rather different. The Phoenix drive (top) is older than the Elpis drive on the bottom. Production dates for drives point to April for the older product and June for the newer. A previous version of this post misstated the dating, ET regrets the error. Thanks to Eldakka for catching it.
And — just as we’ve seen from Crucial and Western Digital — performance in some benchmarks after the swap is just fine, while other benchmarks crater. Here’s what write performance looks like when measured over much of the drive(s):
The original 970 Plus starts with solid performance and holds it for the entire 200GB test. The right-hand SSD is even faster than the OG 970 Plus until we hit the 120GB mark, at which point performance drops to 50 percent of what it was. Real-world file copies also bear this out, with one drive holding 1.58GB/s and one at 830MB/s. TLC hasn’t been swapped for QLC, but the 50 percent performance hit in some tests is as bad as what we see when it has been.
The only thing worse than discovering a vendor is cheating people is discovering that lots of vendors have apparently decided to cheat people. I don’t know what kind of substances got passed around the last time NAND manufacturers threw themselves a summit, but next time there needs to be more ethics and less marijuana. Or maybe there needs to be more ethics and marijuana, but less toluene. I’m open to suggestions, really.
Let me tackle some of the usual excuses out of the gate.
“It’s fine because 800GB/s is still really fast.”
“It’s fine because people don’t copy more than 120GB of data at a time.”
“It’s fine because it’s still TLC NAND, not QLC NAND.”
“It’s fine because everyone else is doing it.”
No, it isn’t.
I’ve already written one article this week with the title “Why Lying About Storage Products is Bad.” The events of the last few weeks beggar belief. Did some silicon Voldemort reactivate a long-dormant “Screw your customers!” sigil in the sky? Are you all being collectively blackmailed? Is the Illuminati involved?
When NAND first hit the consumer market, there were a lot of concerns around longevity and product quality. While there were exceptions, the majority of consumer SSDs from major companies over the last decade have been solid drives. Issues with drive manufacturers bait-and-switching customers like we’ve seen this week have occurred, but they’ve typically been limited to a single company at once. Samsung has been a major player in the NAND market from the start and has enjoyed a good reputation with the general community.
The pandemic is no excuse for this. The pandemic didn’t force Samsung to write “970 EVO Plus” on both products. These products are not equivalent. They are not fungible to anyone who wants to copy more than 120GB of data.
A 120GB cache may seem like it’s enough for most cases, but it’s not a real cache — it’s repurposed/empty TLC (or QLC) NAND that’s being treated as an SLC write buffer. The less free space on your drive, the less cache you have. The drive can’t provide you with 120GB of SLC without roughly 360GB of empty TLC (or 480GB of QLC) and how much cache it can provide you at any given moment will depend on how much you’ve recently written to the drive, how recently it performed garbage collection/TRIM, and how much free space you have. That 120GB SLC cache is at its largest and fastest when the drive is new and mostly empty. As the drive fills and approaches 2/3 capacity, the total amount of cache and overall drive performance will both begin to drop.
One more time, for the companies at the back of the room who may or may not be eating paste: When you ship one product to reviewers and another to customers, you damage trust between the review community and yourselves, between websites and their readership, and between readers and yourselves. You encourage people to assume the worst about your willingness to put profit ahead of the customer experience. You erode the willingness of the community to take your performance promises on faith.
Do you want the launch of future Samsung products to be met with jeers of “Yeah, but how will it perform a year from now?”
Do you think it’s good for Samsung’s brand loyalty if users are forced to play detective in order to make certain they aren’t buying the inferior version of your product that you refuse to identify as such?
The product is not the problem. The lying by omission, lost performance, and dishonorable behavior are the problem.
A few days ago, I recommended both Intel and Samsung as alternatives to Crucial and WD. WD has since pledged to help affected customers and Samsung got caught as well, so I’ve reached out to Intel to make certain they are not also engaging in this process, and to a number of other SSD manufacturers as well. We will report back with what we hear. Evidently there’s been a fair number of component swaps like this rumbling around the storage market in general.