Samsung 860/870 SSDs Continue Causing Problems For Linux Users

[X_m7]

I own a couple of 860 SSDs. What am I supposed to do now, enable a periodic TRIM service (or something similar)?

This whole queued TRIM malarkey is for the continuous TRIM method, which is enabled if you see "discard" in /etc/fstab. Without queuing support, it can cause the system to freeze often, so yes you should switch to periodic TRIM ("sudo systemctl enable fstrim.timer" should do the trick). Apparently Ubuntu 18.04 onwards at least already use periodic TRIM by default, so you might not even need to worry about it.

Source: Glorious Arch Wiki

[rogerx]

No problems here with a Samsung 870 SSD and using periodic TRIM.

The previous comment makes more sense!

Nothing like creating a little panic among the Samsung SSD owner crowd.

Hope you are all prepared for the rioting penguins.

[brianstamper]

So basically this only matters if you insist on using continuous TRIM or have an ATI controller. Most could/should just do the weekly TRIM with fstrim.timer instead in which case queuing doesn't matter all that much. If you do have an ATI controller you will be without TRIM all together, which reportedly could have some performance implications. But, otherwise, saying "Linux users are best off just trying to avoid the Samsung 860 and Samsung 870 series drives" is a bit of an exaggeration.

[LeJimster]

Hmm,

I use an 850 Evo and an 860 Evo and I am not aware of any problems.
How can I find out, if my drives/controller will bring issues to me?

I use a B350 Motherboard, but MSI has no Info, what exact controller is in use.

I have a B350 mobo with 3 Samsung SSD's.
256GB 960 EVO nvme drive
1TB 850 EVO sata SSD
4TB 850 EVO sata SSD

I haven't noticed any problems during the 4+ years I've been using these drives.. So I'm not really sure what is going on. I would really prefer they either fixed the issue properly rather than gimping those of us who don't experience issues.

[LeJimster]

This whole queued TRIM malarkey is for the continuous TRIM method, which is enabled if you see "discard" in /etc/fstab. Without queuing support, it can cause the system to freeze often, so yes you should switch to periodic TRIM ("sudo systemctl enable fstrim.timer" should do the trick). Apparently Ubuntu 18.04 onwards at least already use periodic TRIM by default, so you might not even need to worry about it.

Source: Glorious Arch Wiki

Ok, that makes more sense. It seems I'm already using periodic trim on a weekly schedule when I analysed my journalctl for fstrim. It has been so long since I set up the SSD's to use trim I don't remember these details.

[Guest]

What's all the details about the 850? Is NCQ disabled with it entirely on linux? Is it only with AMD, or specifically X470? Does disabled queue trim affect doing periodic trims with fstrim?

I dual-boot Linux nowadays, and I have it installed to a 850. Prior to this, I wasn't aware of any issues (none popped up, not that I was looking for any).

Edit: https://git.kernel.org/pub/scm/linux...a-core.c#n3971 has ATA_HORKAGE_NO_NCQ_TRIM and ATA_HORKAGE_ZERO_AFTER_TRIM.

It sounds like ATA_HORKAGE_ZERO_AFTER_TRIM means the SSD either doesn't actually zero cells after trim, or it doesn't report that it zeros cells. There's discussion here about the descriptions it seems: https://patchwork.ozlabs.org/[email protected]/

[polarathene]

Crucial, Samsung, and others are known to switch out ICs after review release to cheaper parts. Who knows how long they've been doing this.

This is not an uncommon practice in tech, it's been well known for other products like Wifi dongles/routers or USB flash sticks or external storage too.

Most of the time the average consumer won't even know the difference, especially on Windows. On Linux it can be more of an issue, especially if a chipset vendor is changed affecting drivers, hard enough at times finding decent hardware compatibility for products like these vs SSDs.

It's also quite common with Laptops, a display is often not even in the product specs beyond the obvious features consumers generally are aware of. You have to purchase the damn thing usually to get that sort of information, which can be important info for knowing how well power management will work on it, especially with Linux.

---

In this case, Samsung retained TLC NAND, and changed the controller to their newer ELPIS one. Extreme Tech writes an article in poor taste, it's focusing on a test that makes the product appear to be worse (as in every way, when that's not the case and in other contexts it is superior).

Extreme Tech and the Tom's Hardware article are both right that the product series should have such notable changes more distinguishable to avoid this kind of backlash, but at least Samsung did update their packaging and properly update documentation/specs.

Retailers selling the product should be informed and know better, but they may not care (I've seen some pretty poor inventory management, including online stores), they just want sales and probably didn't want to risk loss of sales from the original sales (stating out of stock and possibly pointing to the new revision is likely to lose sales vs having "exactly" the product name a consumer is seeking to buy). I wouldn't place all blame on a vendor here, you know retailers will deflect blame on to them but they have a responsibility as well.

A similar benchmark from the Tom's Hardware article notes a 150GB file transfer completing a bit faster than the earlier model, reducing the size will only grow that delta, while increasing it like Extreme Tech references will naturally swing in favor back to the original model. It only focuses on sequential I/O throughput, sustained I/O beyond what most would regularly be doing to matter, in majority of cases the new revision will be faster for sequential transfers.

Random I/O is better, as is latency AFAIK, all conveniently ignored because that's not as news worthy. Just point the finger at something to stir up drama and slam a vendor like Extreme Tech did for views. The only point I agree with on that was the packaging should have rebranded the model, Tom's Hardware definitely reported on the issue better.

More concerning would be the poor Linux performance that's reported for that new controller on 980 Pro which I assume now applies to that 970 EVO Plus. Those articles only focus on the Windows perf drop under more niche contexts, while on Linux it's more wide spread isn't? (Linux is too niche to drum up interest on the matter though, would have been a better opportunity to get Samsung to address the issue..)

[polarathene]

Does disabled queue trim affect doing periodic trims with fstrim?

Continuous TRIM (discard in fstab, note that this is not sufficient on XFS root partition however, see Arch Wiki) issues TRIM with each delete and that can compound if it requires doing some GC internally, if you have lots of deletes going on then it can cause perf issues.

To work around that SATA 3.1 compatible devices (with compatible SATA 3.1 controller chipsets and any USB-SATA bridge intermediary if one is involved for external disk) can offer queued TRIM via NCQ, but as this article points out some firmware is faulty and that can result in performance issues that others have said can stall I/O making a system seem frozen/stuck for a bit.

Due to that, it's safer to not worry about such an optimization AFAIK and most distros default/prefer periodic TRIM instead, which sends the command once at some regular interval (eg weekly), to just perform the same task that's piled up in bulk (ideally during an idle period, if it would have any impact on the UX).

As someone else mentioned citing Wikipedia, apparently only Linux supports this Continuous TRIM approach (no clue if that's outdated info). Doesn't seem like there's any practical advantage for consumers to discard on-demand vs once weekly.

So fstrim periodically seems to be the way to go.

[Keats]

Between this and the recent unannounced&unmarked changes to the 970 EVO, I think it's time for people to stop recommending Samsung SSDs.

[billyswong]

Between this and the recent unannounced&unmarked changes to the 970 EVO, I think it's time for people to stop recommending Samsung SSDs.

But are there any brands with better track record?