SSD Thread (Benchmarks, Brands, News, and Advice)

Yeah, it seems fixed for me & others (I was following the 2 big threads on this over on overclockers & somewhere else: can't remember). The new firmware does 2 things: it's 'recalibrated' it's ability to read old data fast, if that first stage doesn't help then the other thing it does is re-write any slow data it detects. Back up to a stable 500MB/s+ for reading the contents of the drive - no issues seen with slow downs on old data anymore.

 

See:
http://www.overclock.net/t/1507897/...-written-data-in-the-drive/3190#post_25012303

The latest firmware may have masked the problem. But 'fixed'? Those are strong words for anything related to Samsung, imo.

The link above shows how bad the problem can be. When I reported that low level of performance with my 1TB 840 EVO's (two years or more ago now???) there was a lot of skepticism from some here...

And for me, the problem arose just after the return period and in a matter of a (very) few weeks.

 

That's a misleading link you put there tiller, the post you're linking is dated 22nd March, yet the firmware fix didn't come out until April 15th! As far as I'm concerned the problem's fixed, (with the one caveat that I've not been actively reading those threads for the past 2 months), but it's fixed for me & for everybody else it seems.

EDIT: oh, apologies, that's the 22nd March 2016! Will read what he's saying & then ammend this post!

EDIT#2: Right, yes, so he'd left his computer off for 6 months & that resulted in a very very slow 840 Evo. He then updated to the latest firmware & it was still slow, but after performing the Advanced Optimisation function in Magician it brought it back to the normal SATA 2 connection speed (he has it connected to SATA 2). The new firmware may not have had time to correct the drive, if he'd have left the machine powered on it might have rectified itself without him running the Advanced Optimisation tool. His machine/SSD should be fine now as long as he doesn't leave it unused for 6 months again! My understanding is that you shouldn't leave SSD's unused for long periods anyway due to potential data loss.

 

Not misleading at all. The link is March 22, 2016.

Hard to lie on the internet with everyone else looking for obvious mistakes and more eyes/brains focused on that than just trying to stick to the topic.

My original response to Mr.Koala says it all...

 

Hi, yep, I spotted my mistake before your post just now. I've now amended my previous post with a response to what you said.

 

The original SSD 'specs' (obviously not official...) was that data retention was guaranteed for 1 year with no power applied. Six months is peanuts compared to what we should expect.

It is actually irrelevant how long he had it powered off. It was giving him those issues way before he semi-abandoned it (minutes to boot windows 7...).

Myself? I don't power down any stationary system unless I'm forced too.

 

Yes, he didn't have the new firmware installed until 'recently', so that's why it was taking him minutes to boot into Windows 7. I will admit though that even with the new firmware in his case (with his particular unit of 840 Evo) if he leaves it for 6 months unused & powered off he will have the slow read speeds again. Once he then turns it back on then it would start to repair the read speed over a period of hours/days, or if he wanted to get back read speed instantly then he would run the Advanced Optimisation in Magician. I do think his usage is an edge case though, most people use their systems every day and this issue wouldn't appear, or perhaps leave it unused for max a week or two during a vacation or something which I imagine would also not be a problem. For the vast majority, edge usage scenarios aside, the problem is fixed.

 

Indeed, they're circumventing the issue by forcing the user to install additional software and having to run that permanently.

Thing is, had my 840 EVOs in RAID (work needs decent consecutive writes) and, lo and behold, the firmware update doesn't work with that, but switching to ahci, flash and then return to raid did alright. Now, you'd expect the new firmware to do its business and occasionally check for difficult reads plus, if necessary, increase the voltage to a cell. Yet it doesn't do this; the software end does and Magician won't work on raid since its crappily programmed (third party apps can access the drives over raid, so why can't Samsung?).

Anyway, it wasn't much of an issue; the DOS firmware flasher had a tool called 'perf.exe' which does the same thing Magician does (rewrite data). So the procedure was switching to ahci, run perf.exe and switch back to raid. Doing this every month or two and the situation was manageable.

And then ... a luminous idea; ' 2x250GB is getting too small', so ... mSATA+1TB SSD = Samsung, unfortunately. Would've gone with another brand this time, buy an SLC even (<- extinct, sadly) or merely an EVO Pro version, but there was only the 840 EVO mSATA 1TB or the 850 EVO to choose from. Well ... the 850 must have this fixed, right ??!

Spoiler: 2x Samsung 850 EVO 1TB - prior to data refresh

Spoiler: 2x Samsung 850 EVO 1TB - after data refresh

Urgh ... and, obviously, 'perf.exe' refuses to run with the 850s; 'incompatible drive' error. Would return these drives, but can't; no alternative for 1TB mSATA. Solution for the interim is restoring the backup image right after it's been written. Might also adjust the 'perf.exe' tool instead, it must use a silly id check somewhere ...

 

You should be posting this to the overclock.net thread(s).

People think Samsung fixes things. Only thing they do is blow more smoke... ... ...

 

It doesn't work on RAID because the tool goes through a special low-level protocol, which the RAID can not pass though. You can't read how old each page is through the normal SATA protocol.

In other words, you can't make a generic cell refreshing tool and expect it to work on every drive out there. If they didn't program against 850, it doesn't work on 850 (assuming that 850 Evo needs such a tool at all).

 

To respond to the bit of your post that I put in bold - Magician is not required to be run in order for the new firmware to work properly, they're independant of one another. The new firmware does it's refreshing of old data cells & does it's increased efficiency of reading old data all without the presence of Samsung Magician - you only need Magician to install the firmware and then you can easily choose Magician to not run at boot with no detriment.

I'm surprised to see the graphs you're showing that the 850 Evo has the same problem, as I've not heard that anywhere else before, the SSD cells are designed differently to the 840 Evo so I'm surprised to see that slow down - given all the excessive negativity (I think a little chunk is justified) towards Samsung I can't but think the graphs aren't quite real.

 

Just like you think the graphs in link below can't be real too.

See:
http://www.overclock.net/t/1507897/...-written-data-in-the-drive/3190#post_25012303

Btw, this isn't the same problem (technically), but in one sense it is (i.e. the 'problem' is Samsung).

 

What are you talking about, that's the same link you gave a few posts back, we've discussed that before anyway, and I never said that those 840 Evo graphs weren't real.

 

Reposted for your convenience.

The issue with the 840 EVO's is that the read speed on old data got progressively slower and slower.

The issue with the specific 850 EVO's in RAID is that Samsung's firmware and/or the RAID0 driver used isn't passing TRIM to the drives.

That is what I'm talking about...

As for the 840 EVO graphs... you didn't think they were 'really' from 2016 for a few minutes, at least.

 

It seems to me that you're confusing issues and being inaccurate on purpose. First you make the incorrect statement about me not supposedly recognising the slow down problem that affected (and in some very edge usage scenarios still affects) 840 Evos and now you're talking about TRIM on 850 Evos - as far as I know TRIM is related to write speeds, not read speeds which is what we're talking about and which was the subject matter of the supposedly "850 Evo Slow Read Speed Graph" shown in a previous post. It seems to me you're just being nebulous & creating confusion, you're normally far to rigourous & accurate to just be not paying attention to the discussion.

 

TRIM is functioning fine over raid, so that's alright, at least:

Spoiler: 2x Samsung 850 EVO 1TB - trim

That may be true, but while it cannot check the entire file, but it can check each cell individually. So it would be nice to have some non-Samsung raid read speed tests. Can't imagine they'll show the same results and, if these are indeed ok, how do the other manufacturers adjust the cells?

This specific issue notwithstanding, the Samsung drives do offer good performance for relatively few bucks, so they're kinda hard to pass up. Can live with this defect (which it is), assuming a (bi-)monthly refresh is sufficient; have to make a backup anyway and rewriting it right after is only a minor inconvenience.

 

About TRIM: Intel's firmware RAID has supported TRIM for a while. Most drives should work.


About cell refreshing:

It's kinda the opposite. There's no time stamp at cell/block level through SATA. Yes, you can check each cell by reading it, but you still don't know what's the last time it has been written. There is a time stamp at file system level. But accessing a file doesn't mean accessing every page at once, and the time stamp can lie anyway, so it's not useful. There's also metadata to further complicate things.

If Samsung really wants to make a tool that works well though RAID, there might be two ways to do it:
[1] Read all the data and rewrite it blindly, or at least make conservative decisions based on file creation time stamps.
[2] Read all the data section by section. If one section appears to be slower than expected, rewrite it.
At that kind of read speed either will take some time. The latter is also far from reliable.
Accessing the drive's internal time stamps and rewriting only old data is way faster (unless most of your data is old). But they can't do this through RAID.

If I'm tasked with reviving a slow 840 Evo I would just clone the whole drive onto another and clone it back. Guaranteed to work since all cells are rewritten now. But it makes sense for Samsung to provide a faster approach when they can. Would be better if they provided a RAID-compatible plan B though.

If drives from other vendors do show better results, and I'm sure most of them will, that means either the 2bit MLC or SLC cells are better at keeping useful charge and the problem is not showing yet, or their firmware is smarter than Samsung's and can take care of itself by recharging old cells automatically (won't work if the drive was never powered on during a long period of time).


More than one big player in the business are trying to bring QLC NAND to the market now. Holly *** I'm afraid of those things.

 

The 840 Evo latest firmware has fixed the slow read speed of old data, regardless of whether in RAID or not (you just have to take it out of RAID to apply the firmware update), so there's nothing else to be concerned about this particular drive. The one caveat to my previous statement, if you leave the drive disconnected or unpowered for months then the read speed of data could temporarily slow down again until the drive has been left powered on for a 'fairly long' period of time to sort itself out. It's not a big deal, I don't know what all this excessive talk is about. Plus, I'm not convinced that the issue is happening with the 850 Evo, I hold doubt against what was posted by a user a few posts back supposedly showing slow 850 Evos - not seen it anywhere else.

 

Answering some of the following in no particular order...

To be clear; I'm not trying to create confusion or be 'inaccurate'. Sorry if that was the end result.

You are incorrect that leaving an SSD powered off for up to a year or more will affect read speeds in any way (on a proper SSD). Secondly, t456 showed how the read speed was affected on his 850 EVO's (mSATA) and yeah; TRIM is not the issue there. (The SSD Read Speed tester was used, not some other bm utility).

The issue though (as I thought I had made clear before...) is that Samsung doesn't fix problems. They mask them and then ignore. I don't care what we call this new issue with Samsung drives (my issue with Samsung SSD's: always found them 'laggy', eventually) - it all starts with how Samsung tunes their firmware for best 'scores' on some useless synthetic benchmark and in doing so fails to give their drives real world performance that users can actually bank on.

The initial issue with people and the 840 EVO's was also initially reported/documented by a single user. I hope this mSATA/RAID issue of the 850's is looked at more closely now by others as well.

Let's not lose sight of the forest for the trees. Other manufacturers (Intel anyone...) build their SSD's firmware and optimizations with the end user in mind. Samsung builds for their marketing staff and materials...

With regards to TRIM working... that little utility is great. But the final test (always) is real world performance. And there, Samsung falls on it's face (especially when compared to it's highest in class 'scores'). 73MB/s read from an RAID0 SSD array? Welcome to 1990...

The fix for the 840 EVO's is far from optimal. And anyone not running the manual 'advanced' performance optimization through Magician (currently 4.9.5) on a regular basis (I run it monthly (roughly) on a batch of 1TB 840 EVO's) is leaving performance/responsiveness on the table. Not in higher 'bm' 'scores'. But actual real world performance that gets rid of some of the Sammy lagginess that is inherent in all their SSD's I've used.

I want to again point out that leaving an SSD (i.e. a 'normal' SSD) unpowered for months should not and does not affect it's performance. The firmware optimizations that Samsung does is not for the end users ultimate benefit (unless they just want to boast about their 'scores'). Sure, their SSD's are 'good enough'. But anyone that takes the time to compare to other/better options soon finds their sore points. And a mere few dollars saved (possibly) is not enough to get them into any of my current systems (new or used).

 

Well, it's ok, let's see if some other users start reporting issues with the 850 Evo, will be interesting to see how it unfolds. I've just been putting my own views across on the whole 840 Evo experience, because I'm really quite happy with mine, I don't get any lag with mine, seems very snappy in everything it does (as well as maintaining over 500MB/s average read speed on that SSDReadSpeedTester2.04 program that tests for any issues with slow reading of old data). I do appreciate some of the points you made though, it's obviously not the the most rock solid SSD on the market!

 

See:
http://www.tomshardware.com/news/intel-optane-backup-3dxpoint-idf,31609.html


Not worth it's own thread, but nice to see this isn't vapor ware either.

 

Exciting stuff, although they were right in the article that there was no point demonstrating sequential performance (because Intels current products can already match that demonstration), they should have chosen a different way which would show the 4K performance.

 

I ordered Micron M500. I am planning to update firmware on it right after receiving the drive. Anything I should be aware of?

 

Can't recall the details of this purchase?

But what I am always aware of with a new storage device:

I make sure that any little glitch on the system it is used with is explainable to anything other than the HDD/SSD for the first few hours/days/weeks and even the first three months of it's life.

The backup script I follow religiously is adhered to even more fanatically in those first few months.

Before the return period is over... I DO NOT load any of my data or my clients (including sign in accounts, banking, etc.) data on the drive/platform in question. If the platform with the new storage device so much as looks at me sideways (inexplicable by any other component), the drive is returned for a full refund. I may test another sample depending on the timeframe I have to getting the system working... otherwise, I would probably be done with a new SSD model/manufacturer that I was not aware of and was just testing.

Also within the return period; I will 'punish' the component (SSD in this case) as heavily as I could in a non-stop manner - if it fails; it will fail here. Not with bm runs (of course), but with real world workloads that would be some multiple of what I would be doing with the drive in actual use.

I do not 'know' the Micron M500 personally. I would be as wary as a mouse with it.

 

@tilleroftheearth the drive is purchased slightly used at very bargain price. It is so-called "enterprise" version of Crucial M500 - the firmware should be a bit different and that's all I guess. Yes, I know it's not the brightest idea to buy a used high-capacity drive, but I am restless since the day I failed to get a new 1TB Cloudspeed Ascend for amazing price, and this Micron drive should settle me down - if it turns out to be OK. Even used, M500 should be more reliable than TLC crap.

 

I'm very happy with mine so far, FWIW.

 

@tilleroftheearth

Here is a benchmark I took just before installing PerfectDisk 14, then another benchmark after installing PD, running an SSD Optimize, then running a TRIM command on the 950 PRO NVMe 512G SSD



Before PerfectDisk:



After PerfectDisk:

 

10.000 Words on how SSD work.

http://arstechnica.com/information-...revolution-how-solid-state-disks-really-work/

John.

 

Two 240GB PNY CS1311 in RAID 0 on a Dell Latitude E6440.

 

Some nice scores there. Unless you're moving around & editing large files RAID 0 isn't really any faster than a single SSD, 4K performance and real world performance (previous example excluded) are normally slightly decreased when using RAID 0, so for most people RAID 0 doesn't make sense. Hopefully you work with large files, in which case RAID 0 is a good call for you.

 

No, I don't work with large files. I did RAID 0 because why not? Plus it makes me feel all warm and fuzzy inside.
No I don't see or feel any speed difference compared to a single SSD.

 

Haha, yes, well it's not a massive slowdown using RAID 0, small but it's measurable (maybe not casually perceivable).

 

I'm trying different AHCI LPM settings, to lower SSD power consumption. Currently have it set to HIPM+DIPM. Did anyone try running "Lowest"?

 

Can anyone recommend an m.2 nvme SSD with low power consumption? Putting it in a notebook (the stealth) that doesn't have a lot of spare juice.

 

I do not know of any that I could 'recommend'. Buy what is available (with a no restocking fee 100% return policy) and go from there.

What does 'doesn't have a lot of spare juice' mean? And what specific model is your 'the stealth' notebook?

 

After you installed PD14, did you check for any available updates?

Did you disable OptiWrite (and double check after a reboot and simply shutting PD down and launching it again that it was actually off)?

SSD Optimize is, imo, useless. Use SmartPlacement with the following settings instead:




Running TRIM after a PD run is not necessary (btw, how long did you wait after you issued that command). But waiting for at least a good half hour to an hour after PD finishes is (do not touch the mouse, keyboard (or touch screen, if you have one) at all during this time).

(Why is waiting important? Because this is what gives the SSD time to do it background GC routine and TRIM sequence (in a TRIM aware environment). Nothing can 'force' a proper TRIM or GC routine to be run on any SSD I know of; the O/S can only 'suggest' it to the SSD and when it feels it is idle enough, it will get to it).

That is a 'proper' PD run for me.

What you did (assuming you did not do the above...)? Merely add more processes to your system for no reason or benefit.

 

1) I always ensure I have the latest software, in fact, before installing, I ensured to grab the latest installer from Raxco again. There were no updates anyway

2) I didn't disable OptiWrite

I'll try that later

thanks bro

 

Give it a try.

But remember; I don't know if the synthetic 'scores' will increase for you. But in actual use, the system will feel smoother and more responsive.

(Btw, the first time I do this on a drive, I do at least three runs to fully defrag all the files and, I also make sure that System Restore is off, hibernation file is disabled (in an Admin command prompt; 'powercfg -h off') and the pagefile is disabled for these first three runs. Afterwards, I enable the pagefile (only) and set it either to 1024MB (min/max) or to 2048MB (min/max) depending on the workstation I'm on (with the 2GB pagefile for the 'heavy' use workstations). Enabling the pagefile (Win10x64Pro) also provides a noticeable improvement to how responsive the system feels).

 

I know bro, I still remember your old advice.

First thing I do when I install Windows:

1- Disable Windows Defender (as I use my own AV)
2- Disable Error Reporting and its service
3- Disable System Restore (because that piece of garbage never helped me anyway)
4- Disable Hibernation (it's actually disabled entirely from my BIOS (thanks to Prema for adding that option in the BIOS) and I delete the page file
5- Disable auto shut down of storage drives after X amount of time through Advanced Power Management

 

I thought that was supposed to be done on every SSD? Disable hibernation, disable system restore, disable index search, disable page file, since it all causes excessive writes to the SSD?

 

I disable all those too, but I don't think we need to worry about SSD write endurance unless we know we're hammering our SSDs beyond a typical user.

 

1) I use Windows Defender as I want the least impact on my platforms. No other AV I've tried interferes less with the O/S.

2) YES!

3) YES!

4) YES!

5) I do not touch those settings for the majority of my systems.


I have found that disabling Index Search is a good way to cripple the O/S - I leave it enabled. Disabling the pagefile results in a slower responding system - I leave it enabled.

Excessive writes? Nothing to worry about with 33% OP'ing (or without; with SSD's bigger than ~500GB for 'average' users).

The above gives the best balance between responsiveness, performance (sustained, over time) and usable capacity for the user - no matter how he/she uses their SSD.

Changing any of the above will throw the 'balance' out the window, imo, and make my computing life worse.

 

I cannot use Windows Defender because if I do, then I get a delay when launching the start menu after a restart. Only happens if I am using Classic Shell. Adding Classic Shell to the exclusions in WD doesn't solve it and I refuse to use the built in Start Menu of Windows 10, that's why I am forced to use another AV.

 

Classic Shell sounds like a virus to me.

Why do you refuse to use what Win10 has? I don't really notice the Win10 start menu (I live on the desktop, myself) at all.

I simply hit the Win key and type 2 to 4 letters of the program I want to launch (if I haven't created a taskbar shortcut for it already). The start menu merely blinks at me (if I notice it at all...). And you do know that you can customize the start menu too right (if you 'had' to use it)?

 

I know, the thing is, in order to get it to look like what I want, I have to do many things. Then when I want to format or go back to an image before installing all programs/drivers, I have to re-do all this again and it's a pain. That's why I prefer Classic Shell, save config via XML and bam, all settings restored instantly

 

@tilleroftheearth

BTW, Windows Defender is not the lightest. See:

http://chart.av-comparatives.org/chart1.php?chart=chart4&year=2015&month=10&sort=1

 

The Razer Blade Stealth takes m.2 NVME drives only. Battery life is important so the longer the better. I did a bit of research and found that the PM951 seems to have slightly lower power consumption than the SM951 which is lower than the 950PRO.

 

Nah, it is the lightest. In terms of no ads, built in compatibility and the least hassle to the user while providing adequate A/V protection (including a user that uses their brains behind the keyboard/mouse).



Not only is it installed out of the box, it self updates, self scans (with the help of Automatic Maintenance, of course) and I never (ever) have to worry about it interfering with any of my programs.

The above cannot be said of almost all the other A/V 'options' that show as a 'lower impact score' vs. Defender.

 

Good to see Avira has the least performance impact, that's what I happen to use, I've been using it since 2009 on the free version, I find it hardly ever nags me too.

 

I've been using Avast since around then and it's not too bad from the looks of it