840 Evo Performance has dropped by ~50%

[James Forrest]

When I installed my 840 Evo around a month ago the Samsung Benchmark was showing ~50k IOPS for both random read and write.

Its now showing ~30k read and 20k write.

What might have caused such a massive degradation in less than a months (light) use.

Current used capacity is 50GB from a total of 250GB and I have ran the Samsung OS Optimisation tools to configure what I assume optimal settings.

Link to picture of benchmark comparison:
http://postimg.org/image/4r83vkvvl/

Thanks in advance

 

[sidamos]

Did you ever solve this?

 

[Szyrs]

The EVO series uses TLC NAND, more fragile than the much more common MLC. They are still highly reliable but SSDs are rated for a life cycle of read/writes (thousands and thousands) as opposed to Platter drives, which are usually rated for hours of life. As a general rule, the read/write cycle is good for several years but one sure fire way to achieve the maximum number of read/writes in the shortest possible amount of time is to bench the fuck out of it.

It's a little unclear what you said about optimising your drive but you should have done that soon after initialising the drive. You would still then have to optimise the drive in windows, or at least ensure TRIM was functioning.

One SSD optimisation in windows than many suggest is to disable or minimise pagefile. Many people disagree with this as it means your system could crash if you use all of your RAM but if you have had pagefile enabled and have also been frequently maxing out your RAM then that means that you will have been frequently using the pagefile space on your SSD as virtual memory (virtual ram).

It's just a guess but if you have not been using TRIM, or not given it space/time to function, and you have also been benching for lolz and using your ssd as system memory (all on the most fragile NAND technology available) then that may have contributed greatly to your current situation.

The how and why is a great curiosity and one I would also like to hear a factual explanation for but regardless of what it is, you should RMA the drive.

 

[sidamos]

In my case, reading old files and running hdparm (short read test) has slowed down after 3 months of normal use (TRIM enabled, using Linux ext4). I did not do extensive benchmarking.

Copying a 3 GB old file to a new file and then (even after reboot) reading the new file is very fast (500 MB/s), where the old file stays slow (170 MB/s). The filesystem is not fragmented much (0.8%).

After leaving the PC on over night, performance of hdparm and reading the old file gets slower and slower and this is not cured by a reboot, so it's more the time the SSD is on power. For example, yesterday after switching the PC on, hdparm said 380 MB/s and now it says 270 MB/s.

As crazy as it sounds, it looks like the "lower" part of the storage gets slower the longer the SSD is on power.

According to SMART, the SSD is perfectly healthy.

I will now issue explicit TRIM and then leaving the PC at the boot menu for the whole day, in case the SSD wants to do garbage collection without being interrupted...

 

[Szyrs]

Do not defragment an SSD!!!! Immediately disable that option.

I'm not sure I follow what you are saying with regards to an old and new file resulting in different drive speeds but it's possibly a result of having defragmented the drive, or that file becoming corrupt or something? Not sure but it definitely seems odd.

What do you mean, "the lower part of the storage"?

TRIM is a background process, so it requires time (I believe best at idle but I could be wrong) to tick away and do its job. Leaving it on now may help with any immediate issues, or it may not - but I tend to try not to be too hasty with my shut downs on a day to day basis. Amongst other SSDs on different machines, I use a 1TB EVO as an OS drive on my gaming rig. I have had no issues with the drive since I bought it last November.

I am not familiar with the benchmark software that you mentioned but i can recommend CrystalDiskMark and AIDA64...

 

[sidamos]

I did not run defragmentation software, I just asked the file system how much is fragmented.

The lower part means lower sector numbers. I guess those are being used first and the older files are there and this is also being read during the hdparm test (Linux read test tool, I don't have Windows).

This morning, I initiated TRIM and then let it sit in the boot menu for 10 hours, so that it has time to do garbage collection or whatever.

Then I booted (PC was on for 30 hours total) and tested with hdparm:
209 MB/s

Then I did a shutdown and switched the PC totally off at the power supply and waited 5 minutes.

After fresh boot, tested again:
437 MB/s

After 10 minutes on:
384 MB/s

After 20 minutes on:
355 MB/s

Edit:
After 3 hours on:
298 MB/s

 

[mad-max79]

You use AHCI?
Which kind of system? AMD or Intel?

 

[sidamos]

Yes, using AHCI. Mainboard is an AsRock P67 Performance with Intel Core i5.

 

[Szyrs]

So do you not have TRIM on by default? This is quite troubling, I'll do some tests with my evo tonight..

 

[sidamos]

No, TRIM was on, but I could not verify if it was really working, so I did a manually TRIM (to no avail).

Normal benchmarks write fresh data and read it again and so they would not find my problem.

If you have an older large (several GB) file, then please make a copy and measure the time. Then reboot. Then make a copy of the copy and measure again. If the second time is much faster, then that would be the same problem as mine.

 

[Szyrs]

Yeah I should have some game folders or somthing that I can move around, or copy.

So what are you transferring to? I have an ocz vertex 3 with nothing on it and a crucial m4 that I could free up some space on. Which would you prefer? I'll try and match your settings as closely as I can.. I only have 1155 sockets boards and no ram over 1600mhz cl9 but I can do you 2600k, 2500k, 2500, 3470, 3570, 3220T or g1620. Ram 2Gb - 32Gb.. Shittest motherboards I have are Ivy Gigabytes but I'll probs use a sandy asus because that's what the evo is on... I know it probably doesn't matter too much but if I'm gonna do it, I may as well try... I also have to bench a caviar green for someone else on here so this will probably take up my evening...

I've only recently started using unix distros though, I can use Kali if you'd rather it over win7?

 

[sidamos]

I have tested with Sandy Bridge and Ivy Bridge boards.

If you do the test under Windows, I would suggest to just copy the file on the same partition, this should also show the problem, because write speed is very fast with the Evo.

If you are using Linux, these are my tests:
hdparm -t /dev/sdX

Raw read, X being the drive letter.

And to copy a file to nothing (I am copying one 3 GB file):
dd if=filename of=/dev/null bs=1M

My PC is now on for 13 hours and hdparm is down to 243 MB/s...

 

[sidamos]

After Secure Erase, also the write performance dropped to 140 MB/s. I am returning the SSD and now I have a SanDisk Extreme II, which performs very well so far.

 

[noisegate]

I have the exact same problem with a 1TB 840 EVO on OSX 10.9. Trim is enabled, I'm using AHCI, yet read times drop dramatically over time, as much as 65MB/s. A fresh copy of any file will read fine.

 

[gino074]

Does anyone has a follow up to this? I have 2 SSD 840 EVO, one 500GB and one 1TB and both have the same exact problem, read speed on old files drops over "on" time as low as ~60Mb/sec. New written files instead are as fast as advertised read/write speed ~500Mb/sec and so if I start benchmark with any app.
After a reboot the read speed improves a bit, but not much...
This seems to me a common hardware problem of those cheap SSD...
Anyone with those SSD and an always-on PC may do some read test speed on OLD-written data?
One good test for this is using the good old HDTach v3 program. It will show very well the drop on read speed on disk zones where old data is, using the "long bench" option. It works very well on windows 7 in XP compatibility mode.

This seems to me a serious problem on these disks. I have also a 256 840 PRO which does not have the problem (same PC and all SSD connected to INTEL AHCI ports.

 

[Palorim12]

IDK if you guys know this, but the 840 EVO has a feature where the hotter it gets, the drive forces its performance to be lower to prevent damage to the drive.

 

[gino074]

Using crystaldiskinfo the temp for both drives is 30 deg celsius. Also, my EVO disks are often idling as the OS in in my 840 PRO.
EDIT: to illustrate, here is a HDTach bitmap of my 500 EVO:

The last part is the one not already written...

EDIT2: to compare, here is my 840 PRO, same pc, same AHCI Intel ports: