Sabrent Rocket Q 1tb extremely poor write performance

[kn100]

Firstly, a little background

PC specs:
Here are the PCs full specs:

Motherboard: ASRock Fatal1ty B450 Gaming-ITX/ac

CPU: AMD Ryzen 2600

PSU: Seasonic GC-500 80+ Gold 500W PSU

GPU: GeForce® GTX 1660 SUPER StormX OC

** NVME SSD:** Sabrent Rocket Q 1tb drive

When I first got this drive, I was getting the rated 3GB/s reads and 1.9GB/s writes, but now, after 12 months, the drive is performing very strangely. Sometimes, I’ll get the full 1.9GB/s write speeds, but most of the time the performance in various benchmarks puts it below 200MB/s.

The drive is partitioned into 4. Two partitions of roughly 450GB, one formatted as NTFS (Windows 10) and then the other is formatted as btrfs, (Ubuntu). The other partitions are the EFI partitions/whatever else Windows 10 makes.

In both Windows and Linux I get this strange performance profile. The problem is unlikely to be thermals, even with my PC fans pegged at 100% and the drive under benchmark, the temperature reported by the drive never exceeds 45 degrees - and I still see the performance problems there.

I have tried TRIMming the disk, both in Windows and in Linux, and there is more than 50% free in both the Windows and the Linux data partitions, meaning there’s over 50% free across the SSD.

Here is an ATTO run, which shows the strange performance I am getting out of this drive.

Here’s a crystaldiskmark run:

And CrystalDiskInfo:

Any guesses as to why this might be? Possibly a failing SSD?

Strangely, now every time I perform the test on Ubuntu (using KDiskMark) I get 1.7GB/s write speeds as I’d expect. I’m not sure I have any idea what’s going on anymore!

[mehmedbasic]

Sabrent Rocket Q 1tb is a QLC drive so if you are utilizing most of the capacity it can get hairy with the performance.

Every cell has 4 bits and the write test writes large amounts that need to go somewhere. On an empty drive they end in separate cells and all is dandy. On a close to full drive there are not many (if any) empty cells and for every 1 bit you write you actually need to read 3 and write 4.

[kn100]

That’s exactly it though, the drive is less than 50% full overall, and has been trimmed numerous times.

[kn100]

And to add a little more, there’s a suspicion there is some sort of incompatibility between Phison E12s based drives like this one and AMD Ryzen systems causing the Cache portion of the drive to not clear properly:

See https://www.reddit.com/r/buildapc/comments/myt2m6/very_critical_write_speeds_on_my_mp510/?utm_source=share&utm_medium=ios_app&utm_name=iossmf for another example of this poor performance on a very similar drive.

This is what I sent Sabrent, but hopefully it helps here too:

[PhaseLockedLoop]

Not a failing SSD my dude its a cheap SSD it will never ever compared to Samsung’s stuff

What’s happening here is every time you write you must write 4 bits regardless of the data because its a cheap QLC drive. Thus you see your slow down on the randoms its quite normal.

If you want more performance with SSDs like anything else one must pay

When you get above about a 1/4th of the capacity of a qlc drive stuff gets unpredictable.

Ada makes a nice cheap drive. SGX1000 I believe I can’t remember but its far better in performance

Sorry its this one

Its no Samsung but its better than sabrent could make if you think you need a replacement

[Remisc]

Be ware of ADATA XPG SX8200 - they recently got caught putting inferior performing components into those SSDs:

Reports of Adata XPG SX8200 Pro bait and switch

I’ve been recommending/using these based on the various reviews I saw on launch. I seem to remember L1 did a great breakdown on it. Based on this, wonder if its time for a follow up? https://m.hexus.net/tech/news/storage/147069-investigation-uncovers-ssd-maker-bait-switch-practices/

[PhaseLockedLoop]

The performance hit wasn’t all that bad. Not enough to notice in your dailies. I remember this. I’m not so hold a grudge against manufacturers unless its consistent like western digital right now.

This was during covid. Chip shortages were occurring and still are. Nothing is ideal rn

[Remisc]

Yeah, it wasn’t that big of a difference but big enough that IMO it pushes this drive from a price/performance spotlight (mainly due to uncertainty brought by ADATA themselves)

From a similar price bracket I would rather recommend Kingston KC2500.
or, oh the irony: WD SN750 (for better sequential and sustained workloads)

Edit: convenient review with WD/ADATA/Kingston drives: https://www.tomshardware.com/reviews/kingston-kc2500-ssd-review/2

[PhaseLockedLoop]

Western digital SSDs are good. I won’t get angry at them for that product like

Its their SMR and CMR and rpm fiasco consistency that makes them consistently untrustable. For spinning rust go toshiba

[mehmedbasic]

@PhaseLockedLoop

Its no Samsung

The issue with the Sabrent is, in part, QLC flash. A Samsung drive with QLC will have similar bottlenecks. A Samsung drive is not better because it’s Samsung but because it’s using more expensive flash.

[PhaseLockedLoop]

Its actually quite a bit better beyond just that of QLC flash. Samsung has QLC models

Samsung however actually spends the in house money on its own flash. Its own quality control. Writes good firmware. Has IRS own controller. This in turn shows great performance and quality over that of other QLC products because of how Samsung operates and a manages their engineering process. So yes it really is in part because its a Samsung. The name is tied to a good engineering process. God help them in not cutting corners right now they have a great process’s right now

[Argone]

I don’t know what to say. I have sabrent, samsung, adata, and from my use they all are within parity of each other.

[PhaseLockedLoop]

The Samsung shines on the ransoms and long sequentials. They are all decent just don’t expect miracles out of the cheaper products.

If you want the best you generally have to pay for the best. That’s just how the cookie crumbles

[kn100]

I’ve given up caring, but to be clear, there is an enormous performance difference when running this same drive with the same windows install, and same data, in an old intel system compared to my relatively modern AMD system. I literally just transferred the SSD between systems and booted into the same Windows install.

Here’s the AMD run again:

and here’s the Intel run:

The Intel system is an I5-6400t based CPU, so absolutely nothing high end or modern, and yet the performance is much better. I get that QLC drives are garbage, but why is the performance SO much better in an Intel system? That is my question, not why my drive isn’t performing to the standard of a MLC or SLC drive!

[HaaStyleCat]

Well, I read everything above… the only thing I don’t see mention of is drivers.

It appears Windows and Ubuntu are working as you like but some other distro is giving you trouble. Could it be a compatibility issue with a certain OS or its ability to use your drive to it’s best ability due to firmware?

I know it’s a simple and common thing people usually have covered, but sometimes forget.

[Zibob]

So I am leaning to something in your setup is wrong somewhere.

I have a Ryzen system and just ran a Cyrstal Disk Mark test and got very different results.

• CPU: 3700X
• Motherboard: X570 (MSi Gaming Edge Wifi)
• Sabrent Rocket Q 2TB (no partitions)

Edit: it just came to my mind, I don’t have any drivers or managing software beyond windows itself. Nothing came with the drive, there was no downloading drivers, no software for speeding up drives or anything. Just turned off the PC, plugged in the drive rebooted formatted and done.

[anon86748826]

@kn100

When I first got this drive, I was getting the rated 3GB/s reads and 1.9GB/s writes, but now, after 12 months, the drive is performing very strangely.

After reading here, I like to ask if you have updated the UEFI in the time between when the drive performed as expected and now? Either some incompatibility was introduced by an update or some settings changed during the update that have a performance impact. Also have you checked that the M.2 slot on the mainboard is in pristine condition? Have you tried another M.2 slot on the board?