Preliminary results for Crucial BX200 concerning degraded read speeds

[Glaring_Mistake]

So, what can be done if your boot drive is suffering from degredation?
Is the only option to secure erase and reinstall?

Well, if you have an SSD with read speed issues using the same controller as this one (SM2256) then there is an option besides performing a secure erase or running DiskFresh.

And that would be to run benchmarks that read the entire drive because read speeds do improve (they are not necessarily restored but improved) when you’ve read the entire drive.
I think that a suffient amount of reads should pretty much restore read speeds (depending on the condition of the SSD).

 

[cbn]

Well, if you have an SSD with read speed issues using the same controller as this one (SM2256) then there is an option besides performing a secure erase or running DiskFresh.

And that would be to run benchmarks that read the entire drive because read speeds do improve (they are not necessarily restored but improved) when you’ve read the entire drive.
I think that a suffient amount of reads should pretty much restore read speeds (depending on the condition of the SSD).

How about Virus scans?

 

[Glaring_Mistake]

Maybe a full drive virus scan could improve read speeds as well?

The important thing is that the drive (or just the files on the drive) is/are read so that should work too even if I haven't tested that specifically.

 

[hojnikb]

i thought siliconmotion uses rewriting techniques like 840evo ?

StaticDataRefresh feature certainly suggest that.

 

[zlejedi]

So, what can be done if your boot drive is suffering from degredation?
Is the only option to secure erase and reinstall?

Well there's only one proper solution to TLC problems - don't buy them while you can still find remaining leftovers of bargain priced MLC drives.

With only exception to that rule being VNAND TLC drives like 850 evo

 

[pcslookout]

Does the Samsung 850 pro suffer from this problem ?

 

[Glaring_Mistake]

i thought siliconmotion uses rewriting techniques like 840evo ?

StaticDataRefresh feature certainly suggest that.

Read speeds do tend to be improved after performing a read on the drive, not restored but improved and which I think is the LDPC adjusting for the condition of the drive.

I've seen a few old files that have had their read speeds restored which may indicate that they have been rewritten.
But I haven't remembered to compare number of writes between tests or just before and after the test to determine if it rewrites any files or if it is also the result of the LDPC.
Even if so it looks like it is just a select number of files are rewritten.


You can also check my earlier test of the ADATA SP550 here: http://forums.anandtech.com/showthread.php?t=2463086

You can see that even for those areas with really low speeds that it does not seem to rewrite files or at least a significant number of them.

At the time I did think StaticDataRefresh was behind the improvements in read speeds but later I realized that the LDPC may be a better explanation seeing as it did not seem to perform any (or at least many) rewrites.

 

[Glaring_Mistake]

Does the Samsung 850 pro suffer from this problem ?

I'm running tests on the 850 Pro in the unlikely event that it should do so.
But in my opinion, no.
Or at least it should take years of it being left unpowered before it starts to develop issues with read speeds rather than weeks which is the case for the BX200.

 

[malventano]

And even if LDPC ECC is better at adjusting for things like age and temperature than the more common BCH ECC and BX200 can use it to increase read speeds after I have read the entire SSD it doesn't prevent read speeds from dropping.

It is far more likely that this is simply FTL expansion / fragmentation impacting the look-up time of reads. I see this sort of thing all the time in testing. It varies based on the controller speed / firmware efficiency. Back when the X25-M had a slow down issue (that impacted reads when let go too far), that had nothing to do with time / temperature / etc - it was simply a FTL management issue. For more on that, see "Long-term performance analysis of Intel Mainstream SSDs", which I wrote.

 

[malventano]

ATTO disk benchmark would have been better to track this.

If you suspect time-based drift of flash cells, ATTO is the exact opposite of what you need to use to track it, as ATTO lays down a brand new file to run its tests on. ATTO is actually bad for even finding most FTL issues because the file is relatively small and its entries are fresh and will likely remain in the controller cache.

 

[Glaring_Mistake]

It is far more likely that this is simply FTL expansion / fragmentation impacting the look-up time of reads. I see this sort of thing all the time in testing. It varies based on the controller speed / firmware efficiency. Back when the X25-M had a slow down issue (that impacted reads when let go too far), that had nothing to do with time / temperature / etc - it was simply a FTL management issue. For more on that, see "Long-term performance analysis of Intel Mainstream SSDs", which I wrote.

There are a few things I do not see as being in line with that explanation though.

1. Given that no major correction has been performed read speeds are connected to age, the older a file is the slower it is read.

2. Time powered down is also connected to read speeds slowing down.

3. The usage pattern that you in the article say may cause these issues does not really resemble the usage pattern for my BX200.

4.The ADATA SP550 uses the same controller but different NAND and it does slow down but not nearly as fast as the BX200 does.

 

[Hellhammer]

It is far more likely that this is simply FTL expansion / fragmentation impacting the look-up time of reads. I see this sort of thing all the time in testing. It varies based on the controller speed / firmware efficiency. Back when the X25-M had a slow down issue (that impacted reads when let go too far), that had nothing to do with time / temperature / etc - it was simply a FTL management issue. For more on that, see "Long-term performance analysis of Intel Mainstream SSDs", which I wrote.

Hi Allyn!

I don't think so. The X25-M and other older SSDs used binary tree FTLs, which require minimal DRAM, but performance can degrade over time as the lookup time increases as the binary tree grows in size.

Pretty much all SSDs use "flat" FTLs nowadays, which is why they require about 1MB of DRAM per 1GB of NAND, but the lookup time for all LBAs is equal.

 

[Ibn Saeed]

How does with fare with 850 evo compared to 850 pro ? Which one from either of the two, degrades over more ?

 

[Glaring_Mistake]

How does with fare with 850 evo compared to 850 pro ? Which one from either of the two, degrades over more ?

Seeing as EVO uses TLC and Pro uses MLC the 850 Pro should have a clear advantage.

But I've yet to see any issues with either of them.

 

[Ibn Saeed]

Seeing as EVO uses TLC and Pro uses MLC the 850 Pro should have a clear advantage.

But I've yet to see any issues with either of them.

Thanks for the reply.

ill go with the pro, just in case.

 

[BonzaiDuck]

Seeing as EVO uses TLC and Pro uses MLC the 850 Pro should have a clear advantage.

But I've yet to see any issues with either of them.

Could be a common feature for any first purchase of a newer hardware technology. But I leaned to the Pro (my 840) before anything, taking note to the sustainable hit to the wallet, I went for EVO and Crucial MX100 about the same time. I may have damaged the EVO between deployments by terminating a Secure Erase process before completing it. Some say it could be recovered. But this is something completely different from degraded performance and failure over time. It was MY fault.

The only other and identical EVO in the house is the boot-system disk for a splendid i5-3570K system I built from spare parts and minor purchases. It takes a licking and keeps on ticking, but the user wouldn't notice the things mentioned in this thread. I haven't observed anything either . . .

 

[Beer4Me]

Is this a result of the controller or the type of NAND memory used? Can anyone test with with SMI controller based SSD with MLC NAND. All the drives noted so far in this thread are TLC-based.

 

[VirtualLarry]

Is this a result of the controller or the type of NAND memory used? Can anyone test with with SMI controller based SSD with MLC NAND. All the drives noted so far in this thread are TLC-based.

Good point. That could possibly tell us if it's some sort of firmware / controller anomaly, in how it processes the FTL tables and stuff.

 

[Glaring_Mistake]

Thanks for the reply.

ill go with the pro, just in case.

Don't really think you need to worry about the EVO having issues when Trion 100 seems to be holding up well despite it using a lot smaller litography (19nm instead of the 40nm used in the EVO) which means it should be more prone to leaks (than the EVO).

 

[Glaring_Mistake]

Is this a result of the controller or the type of NAND memory used? Can anyone test with with SMI controller based SSD with MLC NAND. All the drives noted so far in this thread are TLC-based.

Given that the SP550 and the BX200 both have this issue and use the same controller but don't slow down at the same rate I think it is likely due to the NAND even if the controller may have an impact too.
Unless one or both of them happen to be outliers.


In regards to testing an SSD with MLC NAND with this controller well, as far as I know the SM2256 has only been used with TLC and not MLC so it is not possible to an apples to apples comparison.

The SSD I'm testing which comes closest to that would be the SanDisk Plus which uses the SM2246XT.
And it is not as tested as often as the SSDs with TLC seeing as the frequency is based on the number of write cycles the SSD has.
Next test of the SanDisk Plus should be sometime next month.

 

[zlejedi]

Watch out which San Disk Plus you get - I've seen some mentions that 480GB variant was switch to TLC and only 120/240 are still MLC.

 

[hojnikb]

Watch out which San Disk Plus you get - I've seen some mentions that 480GB variant was switch to TLC and only 120/240 are still MLC.

any source on that ?

although if i think about it, its entirely possible. The last one i tested had pretty fast random read performance, which would indicate sm2256 being used. i should have tested with a bigger test file, but did bother, since i figured it has to be mlc

 

[Glaring_Mistake]

Watch out which San Disk Plus you get - I've seen some mentions that 480GB variant was switch to TLC and only 120/240 are still MLC.

Hadn't heard that.
Did they switch controller too since the SM2246XT lacks support for TLC NAND and LDPC?

Regardless I got an 120GB so there's no need for me to worry about it using TLC NAND.

 

[hojnikb]

It uses SM2256S ans 15nm TLC flash.
http://www.tweaktown.com/reviews/7726/sandisk-ssd-plus-z410-sata-iii-review/index2.html

Interesting to see SM2256 without DRAM though.

 

[Glaring_Mistake]

Just a bit late to update compared to the last time.

Results of last test:

Read speeds continue to drop but haven't changed that much compared to last time.