Assistance - Calling all Samsung EVO owners

I direct you to my res mon shots of an acronis drive image in process.

The problem isn't direct throughput copying something to the drive or off the drive from clean spare space. That is why crystal diskmark etc doesn't show the issue because that is exactly what these benchmarks do. They create a temporary file in space and read and write to it. That works and isn't the problem.

The problem comes when you are reading data off the drive that has been there for some time.

As per my drive image by acronis. As per someone's SQL tests earlier in the thread.

That is why HDTune shows the problem because it read's every bit of data off the entire drive. it doesn't create a new file.

Now I could accept that HDTune has a bug but acronis as well? Those bugs go away when you secure erase the drive?

If HDTune was reading drives badly it would likely - 1. Read all drives badly and I have shown my sandisk works fine as does other none EVO drives. 2. Not start reading the drive properly after a secure erase.
 
It's a difficult one to narrow down but I wouldn't rubbish HDTune pro 5.5.

Yes HD Tune 2.55 which was old software anyway had a bug. Fair enough.

HD Tune pro is being used a lot now and we are showing consistent more sensible results.

Unfortunately as I have stated the nature of the problem isn't going to reveal itself with many of the other benchmarks because you are reading and writing from newly created files on to the SSD.

You need some thing that is reading existing data on the drive.

A suggestion that might show the issue -

Try to copy the contents of your OS windows\winsxs folder to another drive.

That is a large file normally and watch the estimated completion time or time it with your mobile phone.

This file is likely to be near the start of your drive as it is installed as part of your OS.

Then once copied to your other drive rename it and copy it back.

This will work if you if you have 2 SSD's with similar performance like I do.

If not, once copied back as renamed try and copy the renamed again to the second drive and compare times.

Intel's I/O bench which is much more complicate won't show this up because again it uses a method of creating a fresh file to test and that masks the issue.


In any of the drives we have shown with issues some parts of the drive are performing okay but other sections are really poor and until you try to read data out of the poor section you won't see an issue.

The reason you are probably not seeing the issues are this -

1.An SSD even with rubbish bandwidth will boot fast. It has a fast response time and windows booting generally loads lots of small files where bandwidth isn't important. Even a bad peforming SSD is going to be faster to respond than a HDD so yeah you won't notice it because it probably won't give a difference of more than 1-2 seconds which would be subjective enough for you not to notice and within a margin of error anyway.

2. The games you are likely loading are not in the bad sections of the drive and even if they are I refer you to number 1. Even with bad bandwidth it's going to appear fast unless your game is loading LARGE files in bad section of the drive.


I have already said just looking at the drive and the feel of the drive is fine. I felt the same until I investigated further with acronis and when I saw the problems with acronis I realised I needed to delve further to find the problem. It's not an easy problem to demonstrate because of the nature of it.

For exactly that reason it's why people have not been RMA'ing the SSD's left right and centre because people aren't "feeling" the problem and average user would never detect the issue or if they did they'd probably just chalk it up to a slow game or a one off.

None the less as everyone else has said - That doesn't mean there isn't an issue. There has clearly been shown to be an issue.

I wish I could show you more benchmarks but I don't have a bad drive any more as I secure erased it and it went back to samsung to test. Now I don't think they will find a fault because they won't load it up with data and use it for a month or two. It might take several weeks for the issues to start to appear. I don't think samsung will have done that test, yet.



Update -

Okay download the free trial of AIDA 64 extreme. Go to tools disk benchmark. From the drop down that says about run a linear read test on your bad drive - set the options at the top to block size 64KB to match HDTune Pro it will be set to automatic by default.

I suspect you'll get the same as HDTune pro.

Sorry I don't have a bad drive any more to test.
 
Last edited:
Well at least we convinced you there was an issue ;)

More AIDA results if you still have a bad drive will help ascert the fact there are issues here guys.

My drives are on the way to samsung but I am monitoring my new drives and as soon as I see drops like we have been seeing I'll post it again.

Your absolutely right Mei this isn't normal and everyone who has posted has shown the same thing. This should never happen to read performance on a drive.

The problem for samsung is going to be figuring out what is causing this over time. It's obviously time related because those of us who have secure erased have hammered their drives a bit and still seeing great performance which is how it should be.
 
Rotor -

I agree the winsxs is full of small file data -

However if you carry out that test using SSD A
Do the same test using SSD B

One turns out faster than the other you cannot deny regardless that performance of drive A is worse than B.

Forget the winsxs test for now as AIDA is serving the purpose of giving us another test to prove HDTune's results.

What do you consider real world performance? It's only real world if you notice it? or have a feeling of slow performance?

Acronis can do a sector copy but I did a file level copy which I think is it's default as cloning a drive is much faster at file level.

Just because the files are small is no excuse for poor performance. Though I know and am aware of what you describe about small files. If you look at people's drives though you are seeing GB's in some cases 10's of GB's of drive space that is showing the 50MB/S performance

Regardless you are forgetting a key fact. If the small files were the cause of the issue. Explain to me why after a secure erase with a cloned copy of the EXACT disk (i.e all the small files are still there and put back) is the drive performing at over 400MB/S across the board where it was previously doing 50MB/S in sections. Acronis image creation has gone from 30mins+ to 10 mins to do exactly the same image?

Sorry mate it doesn't fly. I appreciate that this is a difficult issue to accept when you don't really notice the issue until you look for it but that doesn't mean it's not there.

The issue is real world I think we have proven that but I agree even I didn't "feel" there was a problem when I first encountered it for general drive usage. I did however feel there was a problem when a backup that should take 10 mins suddenly start's taking 30 mins.

Most people who tested and did a secure erase put their data back and are not seeing the poor performance they had before. In my case it was an exact clone. So all the data that was there is back yet performance is fixed so your small file thoery just doesn't stand up, sorry but keep the idea's coming.
 
sadly not the frozen drives are quite common that is just because your BIOS locks the drive as soon as it boots.

Most bios's will release the drive if you pull out the power connector and plug it back in during the bios detection.

This is th established way of releasing the freeze. Some bioses like my HERO will do it for you.
 
rotor said:
It's been fascinating to follow this thread, and I will continue watching as am very curious to see what turns up. My main reason for being skeptical is there must be hundreds of thousands of these drives globally, yet somehow nobody has mentioned any problems. There's also a lot of mixing/matching throughout the thread, which is why I've piped up here and there.

I agree that there is something funny going on with the benchmarks, and why secure erase appears to make a difference, is just bizarre. So I continue to watch...

And by the way, cheers for being a good sport with my devil's advocacy.
Click to expand...

No problem I appreciate any idea's at this stage because someone will probably come up with a possible answer eventually there is a lot of knowledge out there.

I am not an SSD or storage expert by any stretch of the imagination either.

I just don't want to underplay this so that it becomes a minor issue that can be ignored if I can help it.

If something mostly works that I paid for that isn't really good enough unless you are honest and tell me that upfront.

I want it fixed otherwise :-)
 
There is something there wazza but the performance is going up there slightly with CPU speed agreed however we are seeing massive drops in performance which I don't think could be attributed to CPU speed you are seeing a fluctuation of maybe 50MB/s there.

The performance drop we see is nearer 350-400MB/s so yeah you are right CPU activity is having some effect but not nearly enough.

Toytown what happens to your queue depth on that SQL run?

I saw this on my acronis backup where my queue depth got to 4 and my CPU was at 100% when the drive was in bad performance.

Now the same run after a secure erase gives me queue depth of 0.2-0.4 which is normal and 25% cpu load at most.
 
Well that would stand to reason I think so it's a good find but not explaining the drops in performance. We are seeing we can get more performance out of the drive by fixing the clock rate at turbo speeds. Okay that makes sense the CPU will respond faster.

A CPU idling though is giving 425MB/s on the graph above that is no where near the performance degradation we have seen on "bad" drives before the secure erase and a CPU can't get any slower than it's idle speed.

Good find though wazza.

In other news my drive looks like it might be starting to go. I am seeing a drop about 60% in to the drive now that drops to sub 200MB/s that wasn't there 2 days ago. In that time I haven't done anything to drive other than general use.

I'm going to keep monitoring it and if it persists tomorrow I will post a screenshot I don't want to chalk today up as the start of things to come until I am sure it isn't going away. I did run the benchmark several times though and AIDA as well and they both show it.

So the drive has lasted about 1 week roughly after a secure erase if this proves to be true.

Also seeing having rapid enabled is showing an odd spike upwards where the drive has clearly cached some data to RAM as it peaks up to about 600MB/s which is faster than the interface could handle so must be coming from RAM and again AIDA shows it.
 
Same for the CPU usage.

As I suspected my drive seem's to be okay when I ran the HDtune test with the CPU running realbench.

So I thought for a moment you might be on to something Wazza but I re-ran the test on HDtune with no CPU usage and the drop has gone.

That is why I was reluctant to post a screenshot until I was sure it was the problem coming back.

So watch this space I did an AIDA run and I saw it again but I'm not convinced just yet.

Until I see consistent trough's which are more than a blip I'm not posting anything because it could be a red herring.
 
Last edited:
You forgot to change the block size wazza

Go up to the top and options change the block size to 64KB

it's set to 8mb in that screenshot.

HDTune uses 64KB blocks.
 
nothing wrong with that drive to be honest. My drive looks the same with that initial slow tiny drop.

Looks clean and healthy.

Nothing like those that have had performance issues they had several drops to 50MB/S that spanned several GB's of data.
 
Those are interesting and I would be more inclined to trust AIDA to be honest.

I wonder if you had re-run the HDTune Pro you would have seen anything different.

Was rapid enabled? Wondering if some caching is helping to tidy the graph. Wild suggestion.

Definitely see that running prime doesn't help any more than making sure the CPU stays in it's full power state which increases performance.

The "poorly" drives are still very much an issue but we are no closer to figuring out how they get in a "poorly" state but it is obviously happening to every drive. We've tested too many and not one has come back okay.

So question is how long does it take to get "poorly" .

Tested both my drives again today and both showing okay or within what I would expect for a normal drive.

It is interesting that all the poorly drives show an increase in access time and if you had ran the iop's test in hdtune I'm pretty sure that wouldn't have been good either.
 
Hanaro - who are great - got back to me to say Samsung have not yet received the drives as it is a national holiday tomorrow but they will update me if Samsung have anything to say
 
This is my drive on the 9th of sept -

SSD-9-9-14.png


My drive now on the 12th of sept -
SSD-12-9-14.png


This was the new drives Samsung sent me after the secure erase it has been about 10 day so far
 
You are all welcome but thanks to everyone who contributed without us all here we would not have got a fix. I was just one voice so well done everyone. Let's hope we get the fix soon. Well done Samsung for not sweeping it under the carpet.
 
You must log in or register to reply here.
Back
Top Bottom