Bought an SSD that ended up being QLC, should I keep it?

[sam_86314]

TLDR: To any current or former owners of a Crucial P3 Plus or Corsair MP600 Core XT, what do (or did) you think about them?

So, I recently bought a 4TB Fanxiang S660 to use as my game drive in my main system. I chose this drive because I had an excellent experience with the 2TB version.

Unfortunately, while the 2TB drive uses TLC NAND (YMTC 128-layer TLC), the 4TB uses QLC (Micron 176-layer QLC.

I've been led to believe that QLC is objectively terrible compared to TLC and can even be worse than hard drives in some cases.

I did a small test comparing the QLC drive to the RAID0 SATA SSD array that it was supposed to replace. I copied Baldur's Gate 3 from my boot drive (2TB WD SN850X) to the same folder on both drives.

The SATA array (4x 1TB Crucial MX500) started out at around 3.5GB/s and then settled in at around 1.8GB/s until I stopped the run at 50%. That's perfectly in line with what synthetic tests show on it.

...yeah, this pretty much speaks for itself. The QLC drive does 3.5GB/s for a bit before plummeting to around 200MB/s. That's absolutely terrible compared to synthetic results.

I also ran it on a WD SN570 (DRAMless TLC drive) I have in an external enclosure, and that drive settled around 450MB/s.

I know that sustained sequential performance isn't really real-world performance, but I do sometimes shuffle large files around on my system, and the SATA array handles that infinitely better.

The QLC drive performs way better in synthetic tests, but I don't know how relevant these are to real-world usage.

And, of course, there's my concern about the endurance of QLC. The specific drive I got claims 2400TBW, but given that it's virtually identical to the Crucial P3 Plus and Corsair MP600 Core XT (both advertise around 900TBW), I have my doubts. The MX500s may not have the best endurance either, but they're still better than this when combined.

 

[evernessince]

Games are typically pretty read heavy. Unless you are constantly cycling out games in your games library it should be fine in regards to endurance.

That's said performance may be subpar when initially downloading games. Once the QLC drive runs out of cache, it can only go as fast as the QLC can write. 200 MB/s is pretty fast for a QLC drive writing direct to NAND. The Samsung 870 QVO can only get 50 MB/s but then again the 870 QVO takes awhile to run through it's entire cache, something like 280 GB for the 8TB model. If you are quickly running into the direct write speeds on your QLC drive, it may not be ideal for a drive where you'll need to make 120 GB worth of writes on a semi-frequent basis.

 

[bonehead123]

IMHO, your testing methodology is flawed...

Perhaps you should try again, using a single drive vs. a single drive, both drives freshly formatted with zero contents, AND some "real world" usage scenarios... then maybe you could make some reasonably valid conclusions...

Also, copying files from one drive to another is not really an accurate indicator of a drive's overall performance capabilities, sure, it is ONE thing to consider, but certainly is NOT an all-encompassing statistic to use when comparing any 2 drives to each other

 

[80251]

Games are typically pretty read heavy. Unless you are constantly cycling out games in your games library it should be fine in regards to endurance.

That's said performance may be subpar when initially downloading games. Once the QLC drive runs out of cache, it can only go as fast as the QLC can write. 200 MB/s is pretty fast for a QLC drive writing direct to NAND. The Samsung 870 QVO can only get 50 MB/s but then again the 870 QVO takes awhile to run through it's entire cache, something like 280 GB for the 8TB model. If you are quickly running into the direct write speeds on your QLC drive, it may not be ideal for a drive where you'll need to make 120 GB worth of writes on a semi-frequent basis.

I've never seen any tests on the Samsung 870 QVO 4 TiB that indicates it drops to 50MB/s in sustained writes and I've never seen xfer rates that low w/mine either. I've read 160MB/s is the worst sustained write perf. seen w/the Samsung 870 4 TiB. Random reads and writes w/the Samsung 870 QVO 4 TiB are better than any HDD I've ever had. The Samsung 870 QVO 4 TiB also has a 4GB Low Power DDR4 SDRAM cache that is larger than all the caches in all the HDD's I've ever had put together too.

 

[joemama]

Normally you wouldn't run out of cache unless you are writing some large files to it so it mostly doesn't matter.

 

[ir_cow]

Normally you wouldn't run out of cache unless you are writing some large files to it so it mostly doesn't matter.

Yep, unless you're copying stuff around, pretty much doesn't matter for consumers.

 

[evernessince]

I've never seen any tests on the Samsung 870 QVO 4 TiB that indicates it drops to 50MB/s in sustained writes and I've never seen xfer rates that low w/mine either. I've read 160MB/s is the worst sustained write perf. seen w/the Samsung 870 4 TiB. Random reads and writes w/the Samsung 870 QVO 4 TiB are better than any HDD I've ever had. The Samsung 870 QVO 4 TiB also has a 4GB Low Power DDR4 SDRAM cache that is larger than all the caches in all the HDD's I've ever had put together too.

Oh the drive can absolutely get that low after long sustained writes. I've see it when filling up my 8TB 870 QVO drives. There are also reviews bearing this out as well:

Note the difference in the above graph between the 1TB and 4TB models. This is down to difference in cache size as the larger the cache the longer the drive will sustain a higher write rate and thus the higher the average write speed it will have. By extension the actual write speed when the cache is full will be lower than the average figures presented in the chart above.

I've compared the sustained write speed of the 870 QVO 8TB to my Seagate Exos 18TB drives and the Seagate Exos drives average 235 MB's vs the QVO 8TB's 50 MB/s when copying games from a PCIe 4.0 drive (SK Hynix Platinum P41). As I said before though, the 8TB QVO has a larger cache than most QLC drives and is ideal when it comes to cache size as most people will not write more than 280 GB at a time. The 4TB QVO halves the cache size and is probably still ok for most users but as for other QLC drivers the same is not possible to say without a review.

 

[goms]

And, of course, there's my concern about the endurance of QLC. The specific drive I got claims 2400TBW, but given that it's virtually identical to the Crucial P3 Plus and Corsair MP600 Core XT (both advertise around 900TBW), I have my doubts. The MX500s may not have the best endurance either, but they're still better than this when combined.

TBW ratings are not the same as the actual endurance of an SSD. They just represent the amount of writes you can do before your warranty is void. Some brands lowball more than others. A random Chinese brand could offer very high TBW ratings for marketing purposes, it doesn't mean anything if they don't have proper customer support. If your drive fails in 3 years, what are the chances you will really get a replacement from Fanxiang?

Usually NAND flash is durable enough that normal consumers don't have to worry about it. Micron QLC isn't bad, but QLC drives tend to struggle with any workload that spills outside the SLC cache.

 

[thesmokingman]

That sustained write speed is a no for me, it would prove to be excruciating when in a hurry.

 

[80251]

Oh the drive can absolutely get that low after long sustained writes. I've see it when filling up my 8TB 870 QVO drives. There are also reviews bearing this out as well:

View attachment 312616

Note the difference in the above graph between the 1TB and 4TB models. This is down to difference in cache size as the larger the cache the longer the drive will sustain a higher write rate and thus the higher the average write speed it will have. By extension the actual write speed when the cache is full will be lower than the average figures presented in the chart above.

I've compared the sustained write speed of the 870 QVO 8TB to my Seagate Exos 18TB drives and the Seagate Exos drives average 235 MB's vs the QVO 8TB's 50 MB/s when copying games from a PCIe 4.0 drive (SK Hynix Platinum P41). As I said before though, the 8TB QVO has a larger cache than most QLC drives and is ideal when it comes to cache size as most people will not write more than 280 GB at a time. The 4TB QVO halves the cache size and is probably still ok for most users but as for other QLC drivers the same is not possible to say without a review.

Um, the Anandtech data shows the sustained writes dropped to 163MB/s NOT 50 MB/s. I've also NEVER seen it drop that low on my own Samsung 870 QVO.

 

[GerKNG]

as a game drive... sure.
but fanxiang is ali express trash. and for that alone i'd imediately send it back.

 

[evernessince]

Um, the Anandtech data shows the sustained writes dropped to 163MB/s NOT 50 MB/s. I've also NEVER seen it drop that low on my own Samsung 870 QVO.

You did not read my entire comment:

Note the difference in the above graph between the 1TB and 4TB models. This is down to difference in cache size as the larger the cache the longer the drive will sustain a higher write rate and thus the higher the average write speed it will have. By extension the actual write speed when the cache is full will be lower than the average figures presented in the chart above.

My original comment was stating the direct write to NAND speed, not the average.

You've never seen it drop that low? Have you actually ever tested the drive? The vast majority of people do not test their products when they get them. Go ahead and write 1TB of data to your drive and sample the current write speed (not the average, as that won't tell you the rate to which you are writing to NAND).

 

[80251]

You did not read my entire comment:



My original comment was stating the direct write to NAND speed, not the average.

You've never seen it drop that low? Have you actually ever tested the drive? The vast majority of people do not test their products when they get them. Go ahead and write 1TB of data to your drive and sample the current write speed (not the average, as that won't tell you the rate to which you are writing to NAND).

I HAVE tested my Samsung 870 QVO, when I backed up my HDD image file to it and it never dropped to 50 MB/s. Are you saying Anandtech's SSD testing methodology isn't as good as yours?

 

[evernessince]

I HAVE tested my Samsung 870 QVO, when I backed up my HDD image file to it and it never dropped to 50 MB/s. Are you saying Anandtech's SSD testing methodology isn't as good as yours?

No, I'm saying you're mistaking average write speed for direct NAND write speed. There's nothing wrong with Anandtech's methodology, you just don't understand the difference.

Imaging a HDD to an SSD isn't a test, you've bottlenecked yourself by having a HDD in the equation. A HDD is going to be anywhere from 35 MB/s to 235 MB/s read speed on average depending on the performance of the HDD and other factors like file size / quantity and drive fragmentation. Even my high end Exos drives drop down to 35 MB/s when copying a lot of small files (and it's 100% de-fragmented as well). By copying from a HDD it means you are giving your 870 EVO vastly more time for the cache to recover and by extension reducing the likelihood of filling up the cache as compared to an SSD to SSD copy. In my test I used a SK Hynix P41 Platinum specifically to ensure the drive I was copying from was not a bottleneck.

You've also said nothing about the size of said image file either and it may have well been under the cache size to begin with. Hardly a test.

 

[A Computer Guy]

Um, the Anandtech data shows the sustained writes dropped to 163MB/s NOT 50 MB/s. I've also NEVER seen it drop that low on my own Samsung 870 QVO.

I've throttled my QVO 1TB by copying DVD images (about 8GB per image) from USB3 external disk however that is highly specific large files with no fragmentation scenario.
Basically full speed copy. Larger drives with bigger caches will likely need faster sources with large enough files to bottleneck them in a reasonable amount of time and when it happens it becomes a pretty painful file copy from that point forward, enough to pause the copy for a bit to allow the cache to recover before continuing.

 

[A&P211]

Games are typically pretty read heavy. Unless you are constantly cycling out games in your games library it should be fine in regards to endurance.

That's said performance may be subpar when initially downloading games. Once the QLC drive runs out of cache, it can only go as fast as the QLC can write. 200 MB/s is pretty fast for a QLC drive writing direct to NAND. The Samsung 870 QVO can only get 50 MB/s but then again the 870 QVO takes awhile to run through it's entire cache, something like 280 GB for the 8TB model. If you are quickly running into the direct write speeds on your QLC drive, it may not be ideal for a drive where you'll need to make 120 GB worth of writes on a semi-frequent basis.

My 870 qvo does 150-155mb/s, not 50, but the 870 qvo only has a cache of 78gb, even in the 8tb version.
I have tested the 870 QVO, I have an 8tb version in a SATA external drive. The cache in the 1tb version is 42gb, the rest are the same at 78gb.

 

[evernessince]

My 870 qvo does 150-155mb/s, not 50, but the 870 qvo only has a cache of 78gb, even in the 8tb version.
I have tested the 870 QVO, I have an 8tb version in a SATA external drive. The cache in the 1tb version is 42gb, the rest are the same at 78gb.

The actual amount you can write to the disk before running out of cache is going to be higher than the actual cache size unless you fill the cache before it has a chance to recover. For me that was approx 280 GB before performance took a dip. AnandTech's review is showing the drive running out of cache almost instantly while Tom's hardware was able to write 175 GB before the cache ran out. My 280 GB figure might be down to the fact that I used a real world data set of mixed files sizes instead of synthetic data which resulted in the transfer taking longer as compared to synthetic data.

 

[A Computer Guy]

The actual amount you can write to the disk before running out of cache is going to be higher than the actual cache size unless you fill the cache before it has a chance to recover. For me that was approx 280 GB before performance took a dip. AnandTech's review is showing the drive running out of cache almost instantly while Tom's hardware was able to write 175 GB before the cache ran out. My 280 GB figure might be down to the fact that I used a real world data set of mixed files sizes instead of synthetic data which resulted in the transfer taking longer as compared to synthetic data.

Likewise in my scenario it took about 2.5 seasons of Dr House before the cache ran out (about 115GB).

 

[sam_86314]

I've decided that I'm not going to use the drive in my main system, but I'm currently testing it for another use case.

I currently have a 1TB SSD in a USB-C enclosure that I use for portable storage and for use with Ventoy for installing OSes. I'm going to try the 4TB S660 to see how it does.

I've been reading things about how QLC has lackluster data retention while unpowered, but I'm not sure if that's a real issue. I've never experienced this with any flash storage devices I own. Regardless, this use case involves non-critical data.

Currently using robocopy (because Windows Explorer is a useless piece of shit) to copy everything from the 1TB drive to the 4TB (both in USB-C enclosures); it's sustaining around 670MB/s after copying 160GB.

 

[evernessince]

I've decided that I'm not going to use the drive in my main system, but I'm currently testing it for another use case.

I currently have a 1TB SSD in a USB-C enclosure that I use for portable storage and for use with Ventoy for installing OSes. I'm going to try the 4TB S660 to see how it does.

I've been reading things about how QLC has lackluster data retention while unpowered, but I'm not sure if that's a real issue. I've never experienced this with any flash storage devices I own. Regardless, this use case involves non-critical data.

Currently using robocopy (because Windows Explorer is a useless piece of shit) to copy everything from the 1TB drive to the 4TB (both in USB-C enclosures); it's sustaining around 670MB/s after copying 160GB.

JEDEC standards place SSD data retention at 1 year for client SSDs held at 30c and below. For enterprise SSDs it's 3 months at 40c and below.

Unfortunately it's very hard to tell how well an SSD will retain data beyond those specified periods as there is very little in the way of real world testing. At least from my searches I haven't found any good real world tests on SSD data retention in the long term. HDDs typically need to be plugged in a yearly basis to re-magnetize sectors. From the limited information we have it appears that SSDs are at least as good as hard drives in regards to data retention but that's based merely on the JEDEC standard. Then there's also the question of if QLC makes an impact as well. Do you happen to have a link to the reading material you reference? I would greatly appreciate it.

 

[sam_86314]

JEDEC standards place SSD data retention at 1 year for client SSDs held at 30c and below. For enterprise SSDs it's 3 months at 40c and below.

Unfortunately it's very hard to tell how well an SSD will retain data beyond those specified periods as there is very little in the way of real world testing. At least from my searches I haven't found any good real world tests on SSD data retention in the long term. HDDs typically need to be plugged in a yearly basis to re-magnetize sectors. From the limited information we have it appears that SSDs are at least as good as hard drives in regards to data retention but that's based merely on the JEDEC standard. Then there's also the question of if QLC makes an impact as well. Do you happen to have a link to the reading material you reference? I would greatly appreciate it.

I didn't really find any high-quality sources, mostly just Reddit posts, but I had stumbled across this.

https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2017/20170809_FF21_Miladinovic.pdf

 

[evernessince]

I didn't really find any high-quality sources, mostly just Reddit posts, but I had stumbled across this.

https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2017/20170809_FF21_Miladinovic.pdf

Thank you very much for the link, that was a very interesting read. So in essence it does appear that QLC does have more retention issues but the paper discusses the use of a soft read to improve that. It's nice to see someone using radar charts, they are great for visualizing things IMO.

 

[80251]

No, I'm saying you're mistaking average write speed for direct NAND write speed. There's nothing wrong with Anandtech's methodology, you just don't understand the difference.

Imaging a HDD to an SSD isn't a test, you've bottlenecked yourself by having a HDD in the equation. A HDD is going to be anywhere from 35 MB/s to 235 MB/s read speed on average depending on the performance of the HDD and other factors like file size / quantity and drive fragmentation. Even my high end Exos drives drop down to 35 MB/s when copying a lot of small files (and it's 100% de-fragmented as well). By copying from a HDD it means you are giving your 870 EVO vastly more time for the cache to recover and by extension reducing the likelihood of filling up the cache as compared to an SSD to SSD copy. In my test I used a SK Hynix P41 Platinum specifically to ensure the drive I was copying from was not a bottleneck.

You've also said nothing about the size of said image file either and it may have well been under the cache size to begin with. Hardly a test.

The size of the image file in question was nearly 2 TiB, but I see your point.

I'd like to see you explain away anandtech's testing methodology though.

 

[lexluthermiester]

So, I recently bought a 4TB Fanxiang S660 to use as my game drive in my main system. I chose this drive because I had an excellent experience with the 2TB version.

Incidental storage(few writes, lots of reads) is what QLC is good for. The slow down on writes is not something you will encounter much if you're not rotating your game titles in and out of the drive frequently. If you're not using that drive as an OS/Boot drive and only as a games drive, keep it and enjoy!

 

[maxfly]

I've decided that I'm not going to use the drive in my main system, but I'm currently testing it for another use case.

I currently have a 1TB SSD in a USB-C enclosure that I use for portable storage and for use with Ventoy for installing OSes. I'm going to try the 4TB S660 to see how it does.

I've been reading things about how QLC has lackluster data retention while unpowered, but I'm not sure if that's a real issue. I've never experienced this with any flash storage devices I own. Regardless, this use case involves non-critical data.

Currently using robocopy (because Windows Explorer is a useless piece of shit) to copy everything from the 1TB drive to the 4TB (both in USB-C enclosures); it's sustaining around 670MB/s after copying 160GB.

I wouldn't worry too much about long term data retention. I have an mx500 (at least 10yo) that I forgot about 5 or 6 years ago. I stumbled across it while looking for a spare drive a few months ago. Plugged it in and it booted right up, no issues at all. Granted, it's only a single drive but maybe that will help put your mind at ease.