cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
Agreed.
15nm Toshiba or WD/SanDisk TLC (not 16nm Micron TLC or 16nm SK Hynix TLC) has been my own personal standard for lowest end but still acceptable NAND. So if Toshiba 3D QLC can at least match that then I would be interested in it for least a secondary PC.
Last edited:
WhoBeDaPlaya
Diamond Member
- Sep 15, 2000
- 7,415
- 404
- 126
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
I'm guessing that would probably entail SK Hynix 14nm MLC NAND coupled to a SMI controller.
Then perhaps for competitors we would have.....
----1TB SSD using an odd number of Toshiba/WD 3D QLC dies probably with Marvell 881079 controller (this assuming SMI does not have a SATA 6 Gbps supporting 3D QLC).
----1TB SSD using an odd number of Micron 3D TLC dies (either gen 1 or gen 2) coupled to an SMI 2258 controller (or Marvell 881079)
Quoting from Tomshardware: "At Computex 2016, a flash controller manufacturer warned us about DRAMless SSDs shipping with low-endurance planar TLC NAND. 1xnm planar NAND has far less endurance than 3D NAND. The vendor told us that some of the new NAND only has 100 P/E cycles."
And: "At Computex last June, one SSD vendor told us about an OEM 2D TLC SSD that will burn through the rated endurance in a little over a year. The SSD has to last a year because of the notebook's one-year warranty, but anything beyond a year's worth of use is up to the user to fix"
So there's no specific mention of where the NAND came from but I think it's just that they basically use whatever NAND is cheapest (and poorly binned NAND may make for a bargain) rather than the drop in endurance being due to using SK Hynix 14nm TLC.
That would be the Patriot Blast, it's a 120GB drive and according to SMART it is going to be worn out after about 12-15TB.
But I don't really trust the SMART-values.
You see, it uses Phison S10 as a controller (and the Trion 100 and 150 seem likely to use a slightly modified version of that controller).
Now the Trion 100 had some issues with drives dying pretty early after launch which was fixed by a firmware update.
Not sure what the original firmware was but I would guess it was firmware version 11.0 or 11.1 seeing as the firmware version 11.2 is the most current version for the Trion 100.
Moving on to the Trion 150 it did not have those issues with drives dying after a short time of use, its issue was a more innocuous one, because it was just that the endurance count in SMART went down faster than it should.
That was with firmware version 12.2 and updating to 12.3 fixed that.
Moving back to the Patriot Blast it too had issues with drives dying after just a month or two like the Trion 100 which was fixed with updated firmware.
According to Patriot's website firmware update is recommended if the drive is using 11.0, 11.1, 12.0 or 12.1 firmware versions.
That means it also used to have 11.0-11.1 revision firmware which likely like for the Trion 100 caused drives to die early.
Newest firmware version for the Patriot Blast is 12.2 - as in the same firmware version that the Trion 150 had when the SMART-value for endurance dropped too fast.
And, well - I doubt that it's just a coincidence that firmware revisions were similar when the Patriot Blast seemed to display similar issues to Trion 100 and 150.
So, likely thing is that it's just a firmware issue (that will never be fixed).
TBW for the Patriot Blast is pretty low however so if you're expecting even just a decent bit of wear and to also stay within warranty you may want to look for another drive.
TBW for the 120GB drive is just 19.2TB and for the 960 GB drive it's 76.8TB.
Don't know if I like its performance but in terms of being more resistant to voltage drift their 15nm TLC NAND in my tests it looks like it beats Micron's and SK Hynix's 16nm TLC NAND.
A bit difficult to do an apples to apples comparison on that however since the NAND is not the only factor.
Possibly that 16nm TLC NAND from Samsung can be good as well.
If it behaves like their 15nm TLC NAND though it should not be too difficult to get voltage drift under control.
For example the Phison S10 also happens to be capable of adjusting for that even with NAND less resistant to voltage drift (Micron 16nm TLC) as long as the drive is not too worn so it is not that farfetched that it could be done.
Also something interesting is that some results I've seen may indicate that the 840 does in fact move files around but does not do it fast enough to prevent voltage drift from becoming an issue.
That a drive has that kind of function may not be that surprising but it was still interesting to see it happen fast enough (if still slow) to actually be noticed considering its known issues.
WhoBeDaPlaya
Diamond Member
- Sep 15, 2000
- 7,415
- 404
- 126
^ You're referring to the vanilla 840, and not the 840 EVO right?
AFAIK, Samsung never even bothered to issue a fix for 840 drives, and IINM the fix doesn't work quite right for the 840 EVO either.
AFAIK, Samsung never even bothered to issue a fix for 840 drives, and IINM the fix doesn't work quite right for the 840 EVO either.
Would not exclude that, their controllers are pretty popular and they do release them at a fair rate.
Also one thing to consider is that Marvell and Phison both have controllers supporting 3D QLC NAND (or are likely to in Phison's case) and both also have ECC technologies that they claim may increase endurance by up to 300%.
And SMI has NANDXtend which they also claim can do the same so it doesn't seem that odd to find out that they also have (or will soon have) a controller on the market supporting 3D QLC NAND?
Though perhaps it may be a similar story as with the SM2246 and SM2256, where they first released the SM2256 which supported TLC NAND but later added support for TLC NAND for SM2246.
Noone actually used the SM2246 in a drive with TLC NAND to my knowledge however.
Yep, the 840 and not the 840 EVO.
Samsung did eventually release an updated firmware for the 840 (much to my surprise) but it looks like it pretty much just enabled Advanced Performance Optimization for the 840.
According to tests performed over at PCPer it also improved algorithms resulting in improved read speeds but Techspot also tested the new firmware and they did not arrive at the same conclusion.
They did see improvements in read speeds with each drive read but also noticed that it was caused by the drive getting hot (the 840 absolutely loves heat during reads, like I've seen improvements of 100-200MB/s that likely is most to do with the temperature).
The 840 EVO still has noticeable voltage drift if left unpowered for a while but if it sees use once in a while the performance should be pretty good (according to my tests).
They made some further improvements aside from it now rewriting a bit more often, looks like it is now better at compensating for different temperatures during write (more even distribution at least).
And they also fixed the issue with heat during reads (unlike the 840 it did not like heat during reads, in fact it could see drops due to heat that almost match the improvements the 840 got from heat).
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
That is interesting you mention that because Phison S10 did not have LDPC ECC.
So with the firmware fix to 12.3, how does Trion 150's (which uses Toshiba 15nm TLC) endurance look like?
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
cbn said:
OK, I do agree that is likely.
So for Western Digital/SanDisk 3D QLC SATA 6 Gbps SSDs using DRAM buffer we should expect either Marvell 881079 or a SMI controller.
But where does that leave SATA 6 Gbps SSDs using Toshiba 3D QLC? With this Corporation's NAND typically bundled with a Phison controller does that leave only the S11 controller? Or would Phison update the S10 to also work Toshiba 3D QLC?
EDIT: For people reading this thread not familar with Phison SATA 6 Gbps SSD controllers here are same basic specs....
Phison S10 (55nm, quad core, eight channel, DRAM buffer, BCH ECC, 2TB capacity via 32 Chip enablers)
Phison S11 (40nm, single core, dual channel, DRAM-less, LDPC ECC, 1TB capacity via 16 Chip enablers)
Last edited:
Well, endurance itself did not decrease that quickly but Erase Count went down far faster than it should have.
That is a bit difficult to determine since endurance is just measured in two SMART-values for the Trion 150 (Erase Count and Host Writes) and Erase Count just gives you the percentage of wear and not the amount of P/E used.
Looking at other drives endurance using the same NAND it can vary quite a bit, seen as low as 400 P/E (Kingston UV400) and as high as 2000 P/E (Plextor M7V).
But my guess would be around 700-1000 P/E (possibly a bit more) seeing as it is impossible to know what the WA is.
Which reminds me: They said that their 15nm TLC NAND was supposed to have similar (or better) endurance as their 19nm TLC NAND and if so it looks like the Trion 100 has the opposite problem that the Trion 150 had, that instead Erase Count doesn't decrease fast enough.
As far as I know the S11 is their only SATA 6Gbps controller that they currently have that would support 3D QLC NAND and may be the only one that they would use for that.
They may make a controller to replace the S10 since it feels like it would be in need of an update but we will see if that would be in high enough demand to make them feel that it is warranted.
Maybe with 3D TLC NAND, so that you get some faster drives but unsure if they would pair it with 3D QLC NAND.
This is purely conjecture on my part however so take it with a grain of salt.
With the Toshiba A100 and the OCZ TL100 however we have seen them use 15nm TLC NAND along with the S11 which is the dual channel, single core DRAMless controller I was talking about earlier.
Which is why if QLC NAND should perform like 15nm TLC NAND then, well - why shouldn't it also be considered a good combination?
Of course if you want some really big QLC drives however (which is pretty much the only advantage QLC NAND has) a new controller may be needed.
Ok, I realized last night that I've run some tests on two controllers that use NANDEdge (one of which is confirmed that it supports 3D QLC NAND and the other just likely does) but forgot to write anything about it.
First the Marvell 88SS1074 (in the MX300), it did a fairly good job of managing voltage drift there.
Will see how it does after a bit of wear however, there was one drive recently tested that kind of disappointed me on that front (different NAND and controller though).
And then we have the Marvell 88NV1120 (Runeng M1) which is not the best or the worst (maybe the second worst) controller I've seen when it comes to controlling voltage drift.
It is probably one of the oddest however.
You should probably keep in mind though that Marvell controllers may demand a bit of work from the drive's manufacturer to get the best performance out of them and I doubt they really bothered with that for this drive.
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
Yes, With Phison S10 working with Toshiba 15nm TLC (despite lack of LDPC ECC) I would think at the very minimum 3D Toshiba TLC would work.
Question is why hasn't Phison updated the S10 now? Could it be they are waiting for Toshiba to release its 3D NAND? (Since Phison is almost exclusively used with Toshiba NAND and they expect it's usage to be even lower volume than the S11)
Or maybe there will be a new SATA 6 Gbps controller (with dram buffer) from Phison? S12? Though, with this mentioned, it looks like in some way the company is trying to position the PCIe 3.0 x 2 NVMe E8 controller as the company's replacement for high end SATA---> http://www.thessdreview.com/ces-2017/phison-demos-entry-level-nvme-e8-controller-ces-2017-update/
Last edited:
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,225
- 126
I think 1000 P/E is acceptable. 100-150 P/E is NOT. I don't think I own any SSDs that have only 150 P/E, but I'm not 100% sure.
I own some Adata SU800 Ultimate 128GB with 3D TLC (Micron).
Adata SP550 120GB TLC
Silicon Power S55 120GB TLC
And maybe one other TLC drive.
Also, some Kingston UV300 and UV400, I think.
Do any of those have low P/E cycles? I'll make sure to throw it into a lower-end browser-oriented PC, rather than a power-user / gaming PC.
I own some Adata SU800 Ultimate 128GB with 3D TLC (Micron).
Adata SP550 120GB TLC
Silicon Power S55 120GB TLC
And maybe one other TLC drive.
Also, some Kingston UV300 and UV400, I think.
Do any of those have low P/E cycles? I'll make sure to throw it into a lower-end browser-oriented PC, rather than a power-user / gaming PC.
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
Here is the article the quote about the 100 P/E cycle TLC NAND came from:
http://www.tomshardware.com/reviews/hyundai-sapphire-ssd,4948.html
If I am not mistaken this is the P/E rating of the raw NAND. (ie, lifespan prior to any life extension created by a controller's error correcting mechanism. See link and images below for example).
http://www.tomshardware.com/reviews/silicon-motion-sm2256-ssd-preview,4066.html
With that noted, in post #37 Glaring Mistake mentioned your UV400 had 400 P/E but the Plextor M7V had 2000 P/E. This wide variation even though both SSDs have the same Marvell controller and Toshiba 15nm TLC NAND,
I wonder if this difference, in part, is affected by the higher amount of DRAM on the M7V drive? (Plextor M7V has twice the DRAM (256MB and 512MB) I normally see on 128GB and 256GB SSDs)
P.S See quote below on how dram-less can affect a drive's life. So I wonder does having a smaller (rather than larger) DRAM buffer also affect P/E cycles?
http://www.tomshardware.com/reviews/dramless-ssd-roundup,4833.html
Last edited:
Well, it's not just Phison trying to do that, there is a dearth of high-performing SATA 6 Gbps SSDs pretty much everywhere.
Endurance should not be much of an issue if Tomshardware's claim of Toshiba's 3D TLC NAND having an endurance of 1500 to 3000 P/E before ECC is taken into account proves accurate.
Given the claims of vastly increased endurance with both Samsungs and Micron/Intels 3D NAND I'll still take such claims with a grain of salt.
Still Toshiba claims that their XG5 using 3D TLC NAND can match the previous version using 2D MLC NAND in endurance so it is possible.
What might be more of an issue is that the S10 was made for 2D NAND using Floating Gate while the new NAND is 3D and using a Charge Trap instead.
Because 2D NAND with a Floating Gate behaves differently from 3D NAND with a Floating Gate let alone 3D NAND using a Charge Trap.
That does not explain however why they then have managed to release the S11 if an update to the S10 was delayed due to such difficulties.
Like cbn mentioned the UV400 has around 400 P/E, think that it may intended to be around 500 P/E but that there is some sample variance.
Don't know what the UV300 and S55 are rated at seeing as I've never seen anyone test that with an UV300 and Silicon Power just goes with the NAND that is cheapest.
The SP550 either uses SK Hynix 16nm TLC NAND or Micron 16nm TLC NAND.
So with the SP550 using SK Hynix 16nm TLC NAND I've seen around 1000 P/E but with the Intel 540s using the same NAND it was rated at like 600-700 P/E so still some variance there.
The BX200 using Micron 16nm TLC NAND that was rated at like 900 P/E (the SP550 and BX200 have basically the same controller).
The SU800 uses Micron's 3D TLC which with the MX300 is rated at 1500 P/E but the Adata SU800 may have lower binned NAND because in one endurance test it failed after like 700-800 P/E (my estimation based on its earlier amount of wear).
If I understand correctly then according to Anandtech's Billy Tallis it should not be much of a factor, quoting from the comments on the preview of the Maxiotek MK8115: "SSDs that have external DRAM usually don't use it as a write cache for user data; it's just for caching the mapping metadata. A DRAMless controller like MK8115 can do pretty much the same amount of write combining and wear leveling that a mainstream controller can. There may be a slight increase in write amplification from the controller having to flush metadata updates to the flash more often, but mainstream drives don't want to buffer those writes too much either, for the same safety reasons they don't put user data in the DRAM.
Remember that write operations are the only ones that significantly affect drive lifetime. DRAMless controllers need to do more flash reads, but read disturb errors are still too rare for that to matter to drive reliability."
Don't know which one of the two is accurate here.
But I think that the difference between the UV400 and M7V is more likely because they use differently binned NAND.
Plextor claims that the move to LDPC increased the endurance by 33% (rather than a 300% improvement like Marvell claims for NANDEdge) and with that managed to reach 2000 P/E.
Which means that if so then they're using NAND that would otherwise be rated at 1500 P/E, and thus it is even then rated about three to four times higher than the NAND in the UV400.
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
cbn said:
Micron 3D NAND still uses Floating gate so maybe there is a chance we could see that being used if the Toshiba 3D TLC won't work with Phison S10?
Reduced demand for performance SATA 6 Gbps SSDs?
With this mentioned, I'm thinking if NAND gets really cheap in 2018 (or 2019) we might see some folks with older systems wanting a SATA 6 Gbps SSD larger than what the Phison S11 (and its 32MB SRAM) can effectively cover. (EDIT: And even at the lower capacity points the Phison S10 would be better.)
.
Last edited:
It still behaves a bit differently than 2D NAND with Floating Gate but that might be easier to implement.
But there's still two things to consider, that is if they want to pair it with a controller using BCH ECC when according to them the endurance of their 3D TLC NAND would then drop from 1500 P/E down to just 500 P/E.
See:
Even if that was not considered an issue the use of non-Toshiba NAND and Phison controllers is a rare one, as you yourself have pointed out.
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
It still behaves a bit differently than 2D NAND with Floating Gate but that might be easier to implement.
But there's still two things to consider, that is if they want to pair it with a controller using BCH ECC when according to them the endurance of their 3D TLC NAND would then drop from 1500 P/E down to just 500 P/E.
See:
Even if that was not considered an issue the use of non-Toshiba NAND and Phison controllers is a rare one, as you yourself have pointed out.
That is good point, then probably Micron 3D MLC would be the only choice for that controller (if such a combination ever actually happened).
P.S. So far I have not seen Phison use Micron TLC (or 3D TLC) in any SSD so far (not even in the reference designs).
http://www.tomshardware.com/reviews/toshiba-tlc-mlc-micron-mlc-phison-s10,4190.html (Article with Phison S10 reference testing using Toshiba 19nm TLC, 15nm MLC and Micron 16nm MLC NAND)
http://www.tweaktown.com/articles/7...iew-toshiba-bics-tlc-micron-3d-mlc/index.html (Article with Phison S11 reference testing using Toshiba 3D TLC and Micron 3D MLC)
http://www.anandtech.com/show/11236...vs-and-gfs-drives-featuring-3d-tlc-and-3d-mlc (Palit SSDs using the Phison S11 controller)
And of course, there is always the Patriot Blast (Phison S10 which uses Micron 16nm MLC NAND, not the 16nm TLC NAND).
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
A very interesting comment Glaring_Mistake made in the other thread--> https://forums.anandtech.com/thread...-various-manufacturers.2510437/#post-38975132
Intel-Micron 3D QLC using a small lithogragphy?
Intel-Micron 3D QLC using a small lithogragphy?
cbn
Lifer
- Mar 27, 2009
- 12,968
- 221
- 106
I'm thinking if Intel-Micron decide to go with 3D QLC using the 20nm (or 16nm?) lithography they would probably wait until a suitable controller (with both very strong ECC as well as the ability to handle a very large capacity) emerges.
This being more likely than going with a larger lithography for 3D QLC.
This being more likely than going with a larger lithography for 3D QLC.
No, the Patriot Blast uses Micron 16nm TLC NAND and the Phison S10 so I would not exclude the possibility to see Phison S10 being used with their 3D TLC NAND but that doesn't make me consider it a likely combination (just not impossible).
Maybe?
Don't know if that sounds like a good decision but if they do go with such a small lithography even for their 3D QLC NAND I fully understand why it would be rated at a measly 100-150 P/E.
They will likely not mention the lithography used for their QLC NAND but perhaps we'll see more details on the endurance soon since they're planning to release QLC NAND with their Gen 2 3D NAND.
But if so, just exactly what controller would be strong enough to do that?
Ideally it should both be able to keep read speeds under control (so that it does not have to use up precious P/E cycles) and have an ECC strong enough to correct errors and recover even data that is starting to fade.
Noticed something in a PDF: http://www.phison.com/edit55big5/uploadfile/Phison 20160512.pdf
Look at 2016 SSD Landscape, seems that Phison intended the S10 to be used with 3D TLC NAND.
This makes me consider that the S10 likely has support for 3D NAND or can relatively easy be made to support it but that noone has chosen to use it with 3D NAND.
To be fair though it is mostly used with Toshiba NAND which was delayed more than once and now the market doesn't look the same as it did in 2016.
Look at 2016 SSD Landscape, seems that Phison intended the S10 to be used with 3D TLC NAND.
This makes me consider that the S10 likely has support for 3D NAND or can relatively easy be made to support it but that noone has chosen to use it with 3D NAND.
To be fair though it is mostly used with Toshiba NAND which was delayed more than once and now the market doesn't look the same as it did in 2016.
Someone pointed this out to me:
Looks like the margins may be slim enough for QLC NAND that it's going to benefit from being treated with care.
Comparing it to this:
it looks like QLC NAND may have a bit more overlap than 2D TLC NAND.
Keep in mind however both that since these are two different sources it may not be an entirely accurate comparison and that we don't know the litography of the 2D TLC NAND Samsung used in their comparison.
Anyway this further indicates to me that any 3D QLC NAND from Intel/Micron should be of a fairly large lithography even if their 3D TLC NAND is not.
Looks like the margins may be slim enough for QLC NAND that it's going to benefit from being treated with care.
Comparing it to this:
it looks like QLC NAND may have a bit more overlap than 2D TLC NAND.
Keep in mind however both that since these are two different sources it may not be an entirely accurate comparison and that we don't know the litography of the 2D TLC NAND Samsung used in their comparison.
Anyway this further indicates to me that any 3D QLC NAND from Intel/Micron should be of a fairly large lithography even if their 3D TLC NAND is not.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 404
-
Facebook
Twitter