Samsung 850 EVO ReadSpeedTester results (slowdowns yes/no)?

[BSim500]

We all know about the 840/840 EVO series related slowdowns of stale data. Likewise, other 16nm TLC planar drives such as the Crucial BX200 are showing significant reductions in speed after just 8 weeks (with barely a raised eyebrow from tech sites that only bother to benchmark drives on freshly written data):-

The Samsung 850 series is based on 40nm 3D-NAND which supposedly has cured the problem via going back to large process nodes and reduced cell interference / leakage design, but I'm still pretty stunned that no tech site is even attempting to bother doing any serious practical verification. So far only one guy "Glaring_Mistake" here on Anandtech forums is doing any of this stuff (above pic courtesy of his thread).

So to all those 850 EVO owners, what results are you seeing in SSDReadSpeedTester and could you guys possibly post the results pic (especially if you've been using it for more than say 30 weeks)? The utility doesn't write or change any data, it simply reads existing files then creates a graph of read speed vs age of data (in weeks) and saves it as a PNG file in same directory (no personal info or filenames are shown, see above example pic). Hopefully the problem has truly been cured, but I'm just trying to gather actual evidence of that along the lines of "I installed Windows 40 / 50 / 60 weeks ago and here's objective proof there is no slowdown problem", which seems to be remarkably thin on the ground. Thanks.

 

[VirtualLarry]

I put a 500GB Samsung 850 EVO into an AIO PC, installed a fresh Windows 8.1, and put it away. I don't remember how long ago that was, but it was a few months. How long should I wait before pulling it out and running that program on it? Should I wait another 2 months or so?

 

[BSim500]

I put a 500GB Samsung 850 EVO into an AIO PC, installed a fresh Windows 8.1, and put it away. I don't remember how long ago that was, but it was a few months. How long should I wait before pulling it out and running that program on it? Should I wait another 2 months or so?

You can run it whenever you like / it's more convenient (more data points the better). Unpowered data retention is certainly one of the hardest to find benchmarks. I guess I'm looking for both powered (does an 850 EVO slow down over a year of regular daily system drive use) and unpowered (does an 850 EVO slow down when used as say cold storage external backup drive unplugged 99% of the time) input. Thanks!

 

[Glaring_Mistake]

So far only one guy "Glaring_Mistake" here on Anandtech forums is doing any of this stuff

Yeah, and here I am with some information on the 850 EVO too.

First, this old graph may not have been done with SSD Read Speed Tester but it doesn't show any issues with this 850 EVO that has seen some use.

Not sure about the exact age of the files on it at the time but I'd say at least a year.


Now for a result for an 850 EVO with SSD Read Speed Tester after being unpowered for five months (except for when running tests).

 

[BSim500]

Yeah, and here I am with some information on the 850 EVO too.

First, this old graph may not have been done with SSD Read Speed Tester but it doesn't show any issues with this 850 EVO that has seen some use. Not sure about the exact age of the files on it at the time but I'd say at least a year.

Now for a result for an 850 EVO with SSD Read Speed Tester after being unpowered for five months (except for when running tests).

Thanks very much for the response! I always enjoy your threads. That's certainly reassuring if speeds are a solid +500MB/s after being completely unpowered for 5 months. I have an old Crucial MX100 (16nm MLC), and speeds after about 40-50 weeks are still good (+480-550MB/s). Watching some TLC drives on anything sub-20nm though and they drop like a rock. I'm generally interested overall in seeing how 40nm TLC 3D stacks up against 16nm MLC planar for long-term reads, but it's been so damn difficult to get any data on the subject as every tech site only seems to benchmark fresh data whilst community members only seem to do this stuff if they've got a problem / faulty one. Thanks again for the feedback. :thumbsup:

If anyone else has been using 850 EVO for several months and wants to post their read speeds, please feel free to chip in.

 

[Glaring_Mistake]

Thanks very much for the response! I always enjoy your threads. That's certainly reassuring if speeds are a solid +500MB/s after being completely unpowered for 5 months. I have an old Crucial MX100 (16nm MLC), and speeds after about 40-50 weeks are still good (+480-550MB/s). Watching some TLC drives on anything sub-20nm though and they drop like a rock. I'm generally interested overall in seeing how 40nm TLC 3D stacks up against 16nm MLC planar for long-term reads, but it's been so damn difficult to get any data on the subject as every tech site only seems to benchmark fresh data whilst community members only seem to do this stuff if they've got a problem / faulty one. Thanks again for the feedback. :thumbsup:

If anyone else has been using 850 EVO for several months and wants to post their read speeds, please feel free to chip in.

I have an MX100 that is going to be tested next month that will have been powered down about as long as the 850 EVO in the test above (around five months) so that should be a pretty fair comparison.
Any interest in seeing those results?

What TLC drives have you seen going down in speed though?
So far I've only discovered two combinations of NAND and controller that suffer from read speed slowdowns, that being the SM2256 paired with 16nm TLC from either Micron or SK Hynix.
There are some other TLC drives I'm running tests on of course but they have either not shown much indication of slowing down or not been tested long enough to say if they do slow down.

 

[BSim500]

I have an MX100 that is going to be tested next month that will have been powered down about as long as the 850 EVO in the test above (around five months) so that should be a pretty fair comparison. Any interest in seeing those results?

Sure. That'd be great. :thumbsup:

What TLC drives have you seen going down in speed though? So far I've only discovered two combinations of NAND and controller that suffer from read speed slowdowns, that being the SM2256 paired with 16nm TLC from either Micron or SK Hynix. There are some other TLC drives I'm running tests on of course but they have either not shown much indication of slowing down or not been tested long enough to say if they do slow down.

840 and BX200 mostly (one budget brand I can't remember the name of offhand), though I am wondering if the others with different controllers that are not showing the effect are somehow "hiding" their slowdowns by rewriting data in the background more often like the 840 "fix". About the only way of testing for that though is to record I think it's "Background Program Page Count" SMART value (at least whatever the one is that's "writes that aren't coming from the host PC", ie, internal operations of wear levelling / stale data refresh / garbage collection, etc) before unplugging it then seeing if it suddenly shoots upwards upon plugging it in, which would be indicative of silently rewriting it (which AFAIK, doesn't actually change the file created / modified time stamps so may not be picked up on by SSDReadSpeedTester).

 

[hojnikb]

i wonder how 850evo 4tb will perform, given its large capacity and 21nm 3D TLC..

 

[Glaring_Mistake]

Sure. That'd be great. :thumbsup:

I'll try to remember to update with results from the MX100 then.

840 and BX200 mostly (one budget brand I can't remember the name of offhand), though I am wondering if the others with different controllers that are not showing the effect are somehow "hiding" their slowdowns by rewriting data in the background more often like the 840 "fix". About the only way of testing for that though is to record I think it's "Background Program Page Count" SMART value (at least whatever the one is that's "writes that aren't coming from the host PC", ie, internal operations of wear levelling / stale data refresh / garbage collection, etc) before unplugging it then seeing if it suddenly shoots upwards upon plugging it in, which would be indicative of silently rewriting it (which AFAIK, doesn't actually change the file created / modified time stamps so may not be picked up on by SSDReadSpeedTester).

Yeah, the 840/840 EVO have been pretty well established to suffer from it.
The BX200, are you basing it on my results or have you seen anyone else's BX200 slow down?

Regarding rewrites it can be difficult to tell on some drives if it did rewrite any files, for example the Trion 100 just shows wear in percentages of write cycles used meaning that even if it rewrote the entire drive you would not be able to tell by looking at it's SMART-values that it had rewritten anything.

However for a number of them it seems that if they do rewrite files that read speeds still do not drop very fast since I try to minimize their time to idle and they're powered down between tests.

And while I believe the BX200 and the SP550 do perform some rewrites it may not be as much as I previously thought.
But I don't know as much as I would like about the mechanisms of read speeds improving for the two either by rewrites or through some other means.

 

[Coup27]

I ran this on my 850 EVO in March of this year and uploaded the results in the original thread, post 15. In summary, there was no slow down after 3 months.

 

[BSim500]

i wonder how 850evo 4tb will perform, given its large capacity and 21nm 3D TLC..

Wow, I had no idea some 850's were 21nm. Do you know if that's just for that 4TB drive, or is the 21nm tied into the 48-layer process? That's not very inspiring if all future lower capacity Samsung drives are heading back towards 21nm (same 840 non-EVO process).

The BX200, are you basing it on my results or have you seen anyone else's BX200 slow down?

I have seen someone else's BX200 slow down, so I don't think it's just your system. A friend of mine bought one to install all games on but after 3 months is seeing average read speeds of old data drop. In fact, he's annoyed enough that he's changed the way he uses it (from putting all games on immediately after OS install, to just putting the OS on, installing games on a 4TB secondary HDD but caching just the ones he's currently playing onto SSD via SteamMover (so game data is always freshly written)). That's both a pretty neat trick that makes the most of small budget SSD's, but equally absurd that confidence is low enough that people end up changing usage behaviour like that (but only for planar TLC).

I ran this on my 850 EVO in March of this year and uploaded the results in the original thread, post 15. In summary, there was no slow down after 3 months.

Thanks for the feedback Coup27. It looks like 40nm EVO is fairly resistant to it. Now we've just got to find someone with a 4TB EVO to test Samsung's new 21nm process...

 

[Glaring_Mistake]

Wow, I had no idea some 850's were 21nm. Do you know if that's just for that 4TB drive, or is the 21nm tied into the 48-layer process? That's not very inspiring if all future lower capacity Samsung drives are heading back towards 21nm (same 840 non-EVO process).

Have seen comments that the 21nm is true for the BL half pitch but that it doesn't mean that they have shrunk it overall.

Quoting AKH0 on the original page about this on EE Times: "The "21nm" naming refers to the BL half pitch and nothing else. Samsung's 24 tier chip used single patterning BL and hence their gates should be cut every two pillars. Later they switched to double patterning BL and cut the gates every four pillars. "

I have seen someone else's BX200 slow down, so I don't think it's just your system. A friend of mine bought one to install all games on but after 3 months is seeing average read speeds of old data drop.

Well I didn't think it was just my system that was affected but it's good to see examples from others.

Some have thought my results to be irrelevant since the drive has been unpowered and that if it was just connected to a computer all the time it would not have any issues.
This despite that this was exactly Samsung said about 840 EVO, that those 840 EVO's that were affected were so because they had been unpowered.

 

[Hellhammer]

Have seen comments that the 21nm is true for the BL half pitch but that it doesn't mean that they have shrunk it overall.

Quoting AKH0 on the original page about this on EE Times: "The "21nm" naming refers to the BL half pitch and nothing else. Samsung's 24 tier chip used single patterning BL and hence their gates should be cut every two pillars. Later they switched to double patterning BL and cut the gates every four pillars. "

Ditto. 21nm is bitline half-pitch and it is the same as in 32L V-NAND. The 40nm and 3Xnm figures you see around are measuring a different part of the cell. Technically there is no right or wrong here, it's just a matter of context.

 

[Alfhw]

Thanks for this thread. Seems that the 850 Evo doesn't have that issue or in the worst case is very little. :thumbsup:
However, just in case in the future I need to manually refresh a 850 EVO which program can I use with Windows 10-64 bit?
Diskfresh doesn't show compatibility with Win10 on their support page and can't check myself because I haven't install Win10 yet.
Samsung Magician on installation guide says that: "Some features may not be supported for all models". One user told me that "Advanced Performance Restoration (or Optimization?)" is not supported on 850 PRO (he doesn't know about EVO) and I would like to know before buying the EVO.

 

[coercitiv]

Diskfresh doesn't show compatibility with Win10 on their support page and can't check myself because I haven't install Win10 yet.

I used Diskfresh in Win 10 on the OEM version of the 840 EVO, before firmware fix was available (came later for OEM). It worked as expected.

Scratch that, judging by the date on my SSD firmare update files, I used Diskfresh on Win 8.1

Later edit: since my first 2% of 840 Pro really needed a refresh, I just did it under Win 10. Everything seems in order.

 

[Adarzh]

My 500GB 850 EVO is just over 1 year old.

 

[Alfhw]

My 500GB 850 EVO is just over 1 year old.

Does it mean that 850 Evo still has speed degradation but only later? Any expert can please explain? I'm every day more confused...

Alternately does the crucial MX200 (not BX200) have a similar issue?

 

[Glaring_Mistake]

My 500GB 850 EVO is just over 1 year old.

While read speeds for your 850 EVO are low I think that it is due to a mix of small files and something interfering with the benchmark.
You're recommended to turn off unnecessary applications, especially antivirus in order to get as accurate a result as possible.

 

[Glaring_Mistake]

Does it mean that 850 Evo still has speed degradation but only later? Any expert can please explain? I'm every day more confused...

Alternately does the crucial MX200 (not BX200) have a similar issue?

That result looks to be due to the drive having a number of small files (small files take longer to read) and something interfering with the benchmark.

The reason for the different results if you compare mine with the one Adarzh provided is that mine contains files that are all of the same size and that the benchmark was run with as many applications as possible turned off to minimize the risk of one of them affecting the result.

Also, if you look closely you can see that the read speed at two points exceeds 600MB/s which isn't possible for an 850 EVO.
Or that if these results were accurate that read speeds drop way faster (initially at least) for the 850 EVO than for any other drive.
Which would mean that there would be something seriously wrong with the construction of the 850 EVO itself (something that we have seen little indication of).

It would also mean that Samsung's 16nm 2D TLC NAND leaks less than the 40nm 3D TLC NAND despite that going by litography the latter should have a huge advantage.

 

[Coup27]

Adarzh can you upload the full results TSV file for closer analysis?

 

[Adarzh]

While read speeds for your 850 EVO are low I think that it is due to a mix of small files and something interfering with the benchmark.
You're recommended to turn off unnecessary applications, especially antivirus in order to get as accurate a result as possible.

My AV was off. So does majority of the applications. I'm doing the testing on a laptop. While the first test was running the computer was in power saver mode. I repeated the test in High Performance mode also...There i got little increase in read speed to 286mb.

Also the first test was done on the Windows OS partition. I did test once again on the storage partition. The links both are provided below in the link

Adarzh can you upload the full results TSV file for closer analysis?

https://1drv.ms/f/s!Av2H4uCqBrm_gQSNt0Q4ZgDTSxnb

 

[Glaring_Mistake]

My AV was off. So does majority of the applications. I'm doing the testing on a laptop. While the first test was running the computer was in power saver mode. I repeated the test in High Performance mode also...There i got little increase in read speed to 286mb.

Not sure what is affecting your results then but they are not very credible.

Otherwise one would need to accept both that it's read speeds are at times able of exceeding SATA III (by like six times in the first test) and also that they can drop by 100MB/s in less than 24 hours.

Neither of which I feel enclined to do.

 

[BSim500]

Also the first test was done on the Windows OS partition. I did test once again on the storage partition. The links both are provided below in the link

Thanks very much for providing more detail, Adarzh! This is precisely the in depth stuff that's most useful. I honestly don't know what to make of it as far as "The 850 EVO is immune to slowdown" is concerned. Sorting the TSV files by "Age (days)" shows a "loose" trend. Ignoring both small files (which have more variation in read speeds anyway) and aberrations over 550MB/s:-

Most newer than 90-day large files are fast +490-550MB/s

Medium age data of say 90-180 days start to slow down. Eg:-
- 152 days old 736MB file = 415MB/s read
- 157 days old 1.3GB file = 457MB/s read
- 166 days old 365MB file = 430-450MB/s read
- 173 days old 735MB file = 390MB/s read

+180 day old data is noticeably slower vs 0-90 day fresh data. Eg:-
- 193 days old 16x +50MB files = 238-443MB/s (372MB/s mean average)
- 221-224 days old, 14 files from 71MB to 1.54GB = 149-260MB/s read (245MB/s mean average)
- 228 days 7x 120-474MB files = 246-271MB/s read (265MB/s mean average)
- 292 days 5x 205-479MB files = 166-256MB/s read (219MB/s mean average)

- From 180 days onwards, read speeds of larger +100MB files seem to highly fluctuate from 150-380MB/s. Although they were ignored, small sub 10MB file sizes were consistently reading back at 450MB/s when freshly written (but later on dropped to 95-300MB/s towards 180-300 days).

^ This stuff really highlights why I created the thread in the first place - we need a lot more testing, and a lot more input from other 850 EVO users to eliminate sampling error. It's impossible to conclude anything with a single data set from a single rig, and the (apparent) slowdowns aren't as bad as the 840 series (down to 20MB/s in half the time), but if multiple other people with a +6 month old 850 EVO can show the same thing, then the effect may not be as non-existent as claimed even for 40nm. At the very least, it may raise awareness for the need for serious data retention testing on SSD's & USB sticks in general for both "you need to know something before using this for long-term archival" uses and also "Don't put too much faith in synthetics that read back 5 min old data" benchmarking, (or why tech sites blindly repeating marketing claims are no substitute for actual testing). Be interesting to see an 850 EVO vs 850 PRO read speed test on identically aged 6-12 month old data.

 

[Adarzh]

Not sure what is affecting your results then but they are not very credible.

Otherwise one would need to accept both that it's read speeds are at times able of exceeding SATA III (by like six times in the first test) and also that they can drop by 100MB/s in less than 24 hours.

Neither of which I feel enclined to do.

No. I cannot artificial environments and data to test my ssd. This is what was shown to me and i did my best to run the tests. I don't know if this is the my problem or the ssd. For me, as a normal user, this ssd works exactly as it was worked for me when i bought it. My windows and linux boots in exactly the same time and my applications are opening exactly the same time.

 

[Glaring_Mistake]

BSim500, I think the fact that read speeds seem to be able to drop within 24 hours of the file being written indicates that the low read speeds are not due to files being read slower as they age but rather that there is something else affecting the results.