Support NeoGAF

[GuinGuin]

Recently, major SSD vendors Crucial and Western Digital have both been caught swapping out TLC NAND in their consumer SSDs for cheaper but much lower-performance, lower-endurance QLC NAND. Samsung appears to be joining them in the part-swapping corner of shame today, thanks to Chinese Youtuber 潮玩客, who documented a new version of the Samsung 970 Evo Plus using an inferior drive controller.

潮玩客 first tested the old and new 970 Evo Plus drives using CrystalDiskMark, and the drives appeared near-identical. But that's because CrystalDiskMark by default uses a very small 1GiB test size, which clearly is not enough to exhaust either drive's write cache. (We at Ars frequently use CrystalDiskMark—but we select larger test sizes, for precisely this reason.)

In longer tests, both drives decrease sharply in performance as cache fills, which is expected. But while the older drive retains nearly two-thirds of its original performance, the newer version craters to less than a third. We can see this effect not only in artificial benchmarks, but also in large file copies, as seen in 潮玩客's later tests.

Samsung seemingly caught swapping components in its 970 Evo Plus SSDs

The Phoenix controller in Samsung's own product images isn't present on newer drives.

arstechnica.com

Shameful stuff.

 

[ZoukGalaxy]

This article is quite dishonest.

Actually this is what happened: they runned out of the original controller because of a factory being idle, so they put the controller from the 980 PRO and increased the cache to compensate while totally changing the packaging and PRODUCT NUMBER.

You can spot it and it's not as bad at it seems (not shameful WD level):

Yes, Samsung Is Swapping SSD Parts Too

Component swap done right

www.tomshardware.com

While synthetic results showed a significant difference, the two revisions performed similarly in a real-world copy test with a 154GB video file. The old version has a smaller SLC cache, but a higher sustained write performance. Although the new version's SLC cache is 173% bigger, it offers 47% lower sustained write performance. At the end of the day, everything balanced out and the new version finished the copying process just a hairline faster than the old version.

 

[The_Mike]

This is the iPhone situation all over again.

 

[Werewolfgrandma]

Yup at 1gb my sn850 is like 10gbs. At 32,GB it's a smidge under 6gbs. Makes me think it's not a cheap swap one

 

[KyoZz]

Ffs

 

[clem84]

That's pretty scummy. Release something, let people review and benchmark it, then modify it with cheaper parts.

edit: I stand corrected.

 

[CuNi]

Funny how Ars is only saying "After Cache is exhausted" yet "forget" to mention that the old drive had a Cache of 42GB and the new one has a Cache of 115GB.
That means that while the old drive already falls down to 2/3rds of it's write performance after 42GB, the new drive keeps writing happily till you cross the 115GB threshold.

 

[GuinGuin]

Funny how Ars is only saying "After Cache is exhausted" yet "forget" to mention that the old drive had a Cache of 42GB and the new one has a Cache of 115GB.
That means that while the old drive already falls down to 2/3rds of it's write performance after 42GB, the new drive keeps writing happily till you cross the 115GB threshold.

Both of those seem like huge numbers for a "cache" What is used for the cache?

 

[ZoukGalaxy]

That's pretty scummy. Release something, let people review and benchmark it, then modify it with cheaper parts.

That's not exact, the part is not cheaper, they run out of original controller because of a factory being idle, so they put the controller from the 980 PRO and increased the cache to compensate. Check article in post #2.

 

[nightmare-slain]

Both of those seem like huge numbers for a "cache" What is used for the cache?

not 100% sure but maybe most used/accessed data is stored there? or when transfering files it uses that.

 

[GuinGuin]

not 100% sure but maybe most used/accessed data is stored there? or when transfering files it uses that.

I mean what sort of memory is used in the cache? Volitile or non-volatile?

 

[CuNi]

Both of those seem like huge numbers for a "cache" What is used for the cache?

Usually the Storage itself but instead of TLC Chips being used as TLC, they run in SLC mode so that they can be written to faster and only afterwards starts to actually write TLC in TLC mode.

Edit for better example:
It is slower and more work to save 3 bits per cell in a TLC chip than to just save 1 bit per cell in the same TLC chip, so some devices have a certain amount of maximum storage assigned to also double as "cache" but then be addressed as SLC storage instead of it's actual TLC storage to finish the write faster. They afterwards start to re-write the data from the fake SLC Chips and save the data in actual TLC to not waste storage space.

Edit 2:
To reiterate on my previous statement.
It's a bit dishonest to show those results without also pointing out the difference in Cache. Samsung's change has some benefits with drawbacks, which should be talked about in full, not just one of the two.
But nonetheless, the practice of swapping out Hardware which changes behavior without informing the public should and thanks to many people out there is regularly and rightfully called out!