SSD Read Speed Tester

[Berryracer]

http://www.techspot.com/downloads/6712-ssd-read-speed-tester.html


TLC NAND based SSDs have recently shown a trend of having crippled performance when reading old files (2 to 3 months old data for example) that could make read speeds go as low as 50 MB/S, which is about an 80% of a drop of their normal performance that can reach up to around 500 MB/S.


Since synthetic benchmarks perform tests by writing a new fresh temp file and then trying to read it, this deterioration in performance is very hard to distinguish due to the nature of how synthetic benchmarks work.


If you have a TLC NAND based SSD such as the Samsung 840, Samsung 840 EVO, SanDisk Ultra II, etc. you may want to run this tool that will read your current data to show you the real world performance when it comes to read speeds to see if you have an issue with your TLC NAND based SSD or not. Samsung recently released a firmware update to fix the issue of the 840 EVO 2.5" (not the mSATA version yet) read speed deterioration by having an algorithm in the SSD's controller that will periodically refresh old data to bring its speeds back to normal.


Here are my 840 EVO mSATA Read Speed results of 2 months old data. I have never experienced this slow down really, either the mSATA doesn't experience this slowdown or I got a very lucky sample as its performance has always been higher than the 2.5" 840 EVO counterpart in my testing:

 

[A.t]

What slows down an SSD is incomprehensible data / lots of small files. Let it be TLC, MLC or even SLC as you can see by my results.

http://img.techpowerup.org/150514/SSDReadspeedtest.jpg

It doesn't have much to do with your drive being a good sample or being mSATA.

While I got 1450mb/s peaks on ATTO when I first set this setup up, I'm not getting the performance of the third drive in the RAID array in this case.

Controllers might also play a role in such cases as with such setups like mine as you might want BBU with write caching on in order to get higher performance though.

 

[Elixer]

What slows down an SSD is incomprehensible data / lots of small files. Let it be TLC, MLC or even SLC as you can see by my results.

Eh? Incomprehensible data?
SSDs could care less what kind of data is written to it. Data is data.
Sure, some controllers are able to compress the data (to get better speeds), and others have dedupe technology, but, these are all real-time engines, and quite fast.
I think sandforce is the only one that compresses the data, and in the case of incompressible data, the hit is pretty marginal.

 

[A.t]

Eh? Incomprehensible data?
SSDs could care less what kind of data is written to it. Data is data.
Sure, some controllers are able to compress the data (to get better speeds), and others have dedupe technology, but, these are all real-time engines, and quite fast.
I think sandforce is the only one that compresses the data, and in the case of incompressible data, the hit is pretty marginal.

Data is not just "data" and SSD's can not "care less about what kind of data written to them".

No SSD will perform the exact same on small files as they do on big ones.

That is why you have this thing called 4K performance, and sequential performance.

 

[Elixer]

No SSD will perform the exact same on small files as they do on big ones.

I never said it did.
This also has to do with the filesystem as well, but this would be getting way OT.

 

[w0ss]

With the firmware update my 1TB 840 evo seems fine. Graph is pretty stable at 500MB/s with a few dips to 300MB/s. I can't say I ever noticed this issue myself it always seemed fast to me.

I have a 500GB 840 non evo that I am curious about but that runs linux so not sure if I can check. Supposedly they are impacted by this but it takes longer to materialize and Samsung says they aren't impacted so there isn't a fix.

 

[Fred B]

Eh? Incomprehensible data?
SSDs could care less what kind of data is written to it. Data is data.
Sure, some controllers are able to compress the data (to get better speeds), and others have dedupe technology, but, these are all real-time engines, and quite fast.
I think sandforce is the only one that compresses the data, and in the case of incompressible data, the hit is pretty marginal.

Got higher results when placing certain binary files on ssd and when doing a copy with such a file it exceed max throughput. How that is possible i do not know

Max read of array XM-25 is 800MB and it show 2200MB
http://imgur.com/4sXod4U

 

[Cerb]

TLC NAND based SSDs have recently shown a trend of having crippled performance when reading old files (2 to 3 months old data for example) that could make read speeds go as low as 50 MB/S, which is about an 80% of a drop of their normal performance that can reach up to around 500 MB/S.

Really? Got any reports of this from any drives not of the 840 series from Samsung, so it can be pinned on TLC technology?

 

[Berryracer]

Really? Got any reports of this from any drives not of the 840 series from Samsung, so it can be pinned on TLC technology?

Not reports, but I know for a fact that Tim (the owner of MajorGeeks.com) who has a vanillla 840 has complained to me about this and is annoyed that Samsung forgot about this drive and didn't release a fix.

Also, is a thread from OC Forums:

http://www.overclock.net/t/1519058/the-we-want-our-840-non-evo-ssd-fixed-also

 

[linster]

My Non-EVO Non-Pro 840 has the read degradation issue. Samsung is just completely ignoring the regular 840, no plans as far as i know to bring the EVO 840 "fix" to the regular 840. The final "fix" for the 840 EVO is just a firmware that periodically refreshes your old data. Of course this will put wear and tear on your drive. But since most users don't come close to wearing out the drive, it should be fine.

However, I'm surprised there's not more of an uproar. The read degradation is due to the TLC cell charge decay and Samsung is basically saying there's nothing they can do about it except to refresh the data once in a while. My question is didn't they know the rate of the decay? There's just no way that they didn't. And if so, isn't it just plain deceitful at worst and questionable at best to throw TLC drives out there and sell them with the read/write specifications that they put down? Cuz the specs are pretty much lying after the data has been around a few weeks.

I can't but compare this to the nVidia 970 ROP/memory controller subsystem spec issue. There appears to have been much more of an uproar over that incident. And that incident was just nVidia not being truthful on why the 970 performed less than the 980. There was no performance degradation, just not disclosing part of what made the 970 slower than the 980. In the case of the 840, there was a huge performance degradation that essentially made the drive completely different from what the specs say you should expect. I think scenario two is much much worse.

 

[Berryracer]

My Non-EVO Non-Pro 840 has the read degradation issue. Samsung is just completely ignoring the regular 840, no plans as far as i know to bring the EVO 840 "fix" to the regular 840. The final "fix" for the 840 EVO is just a firmware that periodically refreshes your old data. Of course this will put wear and tear on your drive. But since most users don't come close to wearing out the drive, it should be fine.

However, I'm surprised there's not more of an uproar. The read degradation is due to the TLC cell charge decay and Samsung is basically saying there's nothing they can do about it except to refresh the data once in a while. My question is didn't they know the rate of the decay? There's just no way that they didn't. And if so, isn't it just plain deceitful at worst and questionable at best to throw TLC drives out there and sell them with the read/write specifications that they put down? Cuz the specs are pretty much lying after the data has been around a few weeks.

I can't but compare this to the nVidia 970 ROP/memory controller subsystem spec issue. There appears to have been much more of an uproar over that incident. And that incident was just nVidia not being truthful on why the 970 performed less than the 980. There was no performance degradation, just not disclosing part of what made the 970 slower than the 980. In the case of the 840, there was a huge performance degradation that essentially made the drive completely different from what the specs say you should expect. I think scenario two is much much worse.

Thank you so much for this post, there's more proof to you Cerb

 

[mikeymikec]

Thank you so much for this post, there's more proof to you Cerb

Read cerb's and linster's posts again.

 

[linster]

Yeah, I think he's looking for a non-Samsung TLC drive to show that it's a TLC issue and not a Samsung issue. The Anandtech post regarding the fix says it is a 19 nm TLC signal degradation issue. Not sure if this is official.

 

[Cerb]

My Non-EVO Non-Pro 840 has the read degradation issue. Samsung is just completely ignoring the regular 840, no plans as far as i know to bring the EVO 840 "fix" to the regular 840. The final "fix" for the 840 EVO is just a firmware that periodically refreshes your old data. Of course this will put wear and tear on your drive. But since most users don't come close to wearing out the drive, it should be fine.

However, I'm surprised there's not more of an uproar. The read degradation is due to the TLC cell charge decay and Samsung is basically saying there's nothing they can do about it except to refresh the data once in a while. My question is didn't they know the rate of the decay? There's just no way that they didn't. And if so, isn't it just plain deceitful at worst and questionable at best to throw TLC drives out there and sell them with the read/write specifications that they put down? Cuz the specs are pretty much lying after the data has been around a few weeks.

I can't but compare this to the nVidia 970 ROP/memory controller subsystem spec issue. There appears to have been much more of an uproar over that incident. And that incident was just nVidia not being truthful on why the 970 performed less than the 980. There was no performance degradation, just not disclosing part of what made the 970 slower than the 980. In the case of the 840, there was a huge performance degradation that essentially made the drive completely different from what the specs say you should expect. I think scenario two is much much worse.

Were it due to normal decay, there would be more problems, and lots of corrupted/lost data, rather than just a speed decrease. There are a few other hypothesis, of which their TLC being sensitive to write disturbs makes the most sense to me (could be wrong, though), and could be something that could slip by them. Some drives might also be stuck running warmer than others, increaseing degradation by more than they'd like. If that TLC just plain degrades too fast, what of other devices with their TLC NAND installed?

No doubt Samsung dropped the ball. They've been showing that they are as good as Sony, lately, about being able to do that.

 

[Dujith]

Since synthetic benchmarks perfor..., just something i'm genuinely curious about.

 

[Berryracer]

Let me start of to say that i have no idea how they actually test and whats under the hood. But i have a 840 EVO with the bug and a fresh install of hdtune showed me the slow speeds at once when i first noticed slow loading times in games.

So if the bench works like you said i would have had no problem shown in the bench, but it did show the very low read speeds. How come?

ps. This is in no way an attack, just something i'm genuinely curious about.

logically speaking, AS SSD benchmark can now show you the TLC performance degradation of Read speeds since it's writing / reading from a new temp file it creates, if there is a slow down even in AS SSD, then the issue lies elsewhere

what driver are you using? AHCI or IRST? if IRST, which one? version please

What is your Antivirus also? which OS?

can you post a screenshot of your AS SSD benchmark? that would also show us if your SSD is properly aligned or not

 

[Dujith]

logically speaking, AS SSD benchmark can now show you the TLC performance degradation of Read speeds since it's writing / reading from a new temp file it creates, if there is a slow down even in AS SSD, then the issue lies elsewhere

what driver are you using? AHCI or IRST? if IRST, which one? version please

What is your Antivirus also? which OS?

can you post a screenshot of your AS SSD benchmark? that would also show us if your SSD is properly aligned or not

The speed issue is already fixed with the evo patch. And like i said i ran HDtune.
AHCI drivers, No antivirus besides Windows Defender and their own anti malware stuff. Windows 7 64bit

If you want i can run the AS SSD benchmark.

 

[Berryracer]

The speed issue is already fixed with the evo patch. And like i said i ran HDtune.
AHCI drivers, No antivirus besides Windows Defender and their own anti malware stuff. Windows 7 64bit

If you want i can run the AS SSD benchmark.

please do and post the screenshot

 

[Dujith]

As you can see 0-7 weeks is top speed while anything older hovers around 380 MB/s with no real pattern to old - very old data
The SSD in question has 60GB free atm

Edit: shoot, this is the techspot one :S i'll do the AS SSD one too
Edit 2: tada

.
Quick google learns that write/read speeds seems okay. Especially in real world usage

 

[Berryracer]

As you can see 0-7 weeks is top speed while anything older hovers around 380 MB/s with no real pattern to old - very old data
The SSD in question has 60GB free atm

Edit: shoot, this is the techspot one :S i'll do the AS SSD one too
Edit 2: tada

.
Quick google learns that write/read speeds seems okay. Especially in real world usage

Here is my older benchmark of my 1TB 840 EVO:

AS SSD Benchmark with IRST 12.0.7.1002 (W7)

Your scores look fine to me but a bit low on the 4K writes

You are using the default MSAHCI driver

so just for comparison, here is the same SSD with the MSAHCI Driver:

AS SSD Benchmark with Intel Chipset Drivers 9.4.0.1023 (W7) [MSAHCI]

I suggest you use the IRST driver is it is a better controller driver vs the ancient MSAHCI Driver which is built for the best compatibility and hasn't been updated since many years

Try and let's see the scores with the IRST driver