Samsung 840 120GB SSD Review @ [H]

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,534
Samsung 840 120GB SSD Review - The 120GB Samsung 840 Series SSD features the powerful 8-channel MDX controller and TLC NAND. While this value SSD comes at a very good price, it also features much lower speeds than its larger capacity brethren. We put this value SSD through our suite of steady state tests to see if it can pass muster.
 
I find them to be pretty decent for virtual machines that have little growth and on older servers that do not have SATA 3. Have not really tested the steady state. Using the Samsung over provisioning of 11 gb kinda sucks. At least when you register on samsung.com they give you 3 more months to warranty.
 
Thanks for the review Kyle. I'm waiting to see what Micron and Intel will be coming out with soon that's based on MLC chips personally, or maybe 840Pro is it drops in price enough. I'm not comfortable with this TLC cost cutting stuff.
 
TLC should last 7 years for the 250GB model, 14 years for 500GB variants, and 3.5 to 4 years for the 128GB model. This is assuming a typical consumer’s 10GB per day workload.

I've seen this many times, and I'm curious about the basis for that 10GB estimate.

Right now one of my most popular jobs involves upgrading older computers (mostly Core2 era laptops at this point) using SSDs. The lower price-point of these TLC drives makes them appealing, especially for those who don't want to spend too much upgrading an older computer.

But for the vast majority of my clients, a typical workload involves surfing the Web, Some light Office 2010 duty (mostly Microsoft Word, some Excel), maybe some Skype, and if they are feeling creative they might print something to PDF. That's about it. That shouldn't come anywhere near 10GB per day. I would think that on most days they probably write less than 1GB to the drive.

One of the newer laptops that I recently upgraded to an SSD is used to play World of Warcraft. The entire WoW install is about 30GB, but once the game is installed there really isn't anything that would continue to write large amounts of data to the drive each day as she played. Even on my main desktop I'm doubtful that I'm doing 10GB of writes per day really.

I just think that it might be worth an investigation, to see what actually constitutes a typical workload. You might argue that with such a light workload, what is the point of an SSD? But users definitely notice and appreciate when their computer boots lightning quick, when apps open fast, and when stuff like a background AV scan doesn't grind the whole laptop to a crawl.

I've already used a few 120GB 840 TLC drives when upgrading people's computers. So far performance seems very good and satisfaction is high, but I'll be sure to keep an eye on the situation.
 
Users must take into consideration the large amounts of random writes that occur constantly in the background for logging purposes, etc. Browser cache/history also can create some residual use that users aren't aware of, and a lot of it depending upon surfing patterns.
The type of write is very important when considering endurance metrics, as random writes create more wear than sequential writes do.
Office automatically saves documents every ten minutes by default (at least with my 2013 install) and many change this to 5 or more. So that document is constantly being saved, etc.
Backup and restore can chew through a ton of space if not disabled, and the hibernation files can be excessively large, used every time the system hibernates this adds up. I typically leave hyberfile.sys deactivated, but reactivated it to take a look at the size. It is 11.9GB. And I haven't even hibernated :).
There are many other programs and system services that require excessive amounts of space and use, even when the user is not aware.
For instance, WoW is logging your trek, your stats and tons of other data. Much of this data is stored in temp files until your session is over, and deleted, all unbeknownst to you.
10GB per day is almost a conservative estimate.
 
10gb per day is a conservative estimate if you are a power user.

File last modified flags, journaling, etc all eat up writes and are often turned off when tuning for SSD use. Hibernation files, page files, logs, windows updates, and program settings all eat up writes. iTunes writes like crazy.

I got an SSD for a work machine. TLC is likely inappropriate for database work that sees any kind of non-trivial write activity. I wish I could afford an SLC drive, but MLC it is.

3.5 to 4 years on a typical user workload is worrying too. I'd consider running one of these in like an HTPC or for virtual machines that don't see much use, but I can't imagine anyone here wanting one for a primary or frequently used drive.
 
I'm using one as my cache drive for a Hitachi 4TB 7200 HDD. Should I be concerned ?
(I use a 256Gb 830 as my C drive)
 
Backup and restore can chew through a ton of space if not disabled, and the hibernation files can be excessively large, used every time the system hibernates this adds up. I typically leave hyberfile.sys deactivated, but reactivated it to take a look at the size. It is 11.9GB. And I haven't even hibernated :).

That is a sparse file. Windows reserves a certain amount, but it only writes as much as needed, depending on what is in RAM when you hibernate, and how well it compresses.
 
I've seen this many times, and I'm curious about the basis for that 10GB estimate.

I've been using an Intel 335 240GB for 24 days now, and I've written 317GB to it so far, for an average of 13GB per day. Like you, I expect this average to decline over time, because most of the data I wrote to was done in the first few days as I installed Win7, apps and games on it.

If you have only installed Windows once at the beginning of your ownership of the SSD, there's an easy way to know exactly how long you've had it: run cmd, and type systeminfo. You'll see the original install date of your operating system. SMART info will tell you how much data you've written to the SSD.
 
I've been using an Intel SSD x25M 160GB since about the time when it came out.

I did the math yesterday, and it has written about 1 GB/hour (Host Write Amount devided by Power On Hours).

The drive has been used for OS, apps, games and everything else you might use your OS drive for. I have in no way tried to spare the drive. It's a harddrive, and if it wears out, I'll get a new one.


Point is, in my case, about 1gbyte of data is written per hour, when the computer is powered on. Stop caring about wearing your SSD's out in non-database usage scenarios. If it was a real problem, we'd have these forums overrun with people complaining about dead NAND cells, but we do not. People complain about crappy firmwares (thanks OCZ) and rightly so.
 
I agree that the majority of users will be fine with standard MLC, definitely nothing to worry about.
Even high capacity TLC will be fine, but there is a cut-off when some solutions will not remain viable for long term use.
 
I wonder how much durability in years does the 840 Pro 256 gb have?
 
Well i"ve a 250gb 840 non-pro as my main harddrive with the games I play most often, notably BF3. Wonder how long it will last
 
I've seen this many times, and I'm curious about the basis for that 10GB estimate.

Right now one of my most popular jobs involves upgrading older computers (mostly Core2 era laptops at this point) using SSDs. The lower price-point of these TLC drives makes them appealing, especially for those who don't want to spend too much upgrading an older computer.

But for the vast majority of my clients, a typical workload involves surfing the Web, Some light Office 2010 duty (mostly Microsoft Word, some Excel), maybe some Skype, and if they are feeling creative they might print something to PDF. That's about it. That shouldn't come anywhere near 10GB per day. I would think that on most days they probably write less than 1GB to the drive.

One of the newer laptops that I recently upgraded to an SSD is used to play World of Warcraft. The entire WoW install is about 30GB, but once the game is installed there really isn't anything that would continue to write large amounts of data to the drive each day as she played. Even on my main desktop I'm doubtful that I'm doing 10GB of writes per day really.

I just think that it might be worth an investigation, to see what actually constitutes a typical workload. You might argue that with such a light workload, what is the point of an SSD? But users definitely notice and appreciate when their computer boots lightning quick, when apps open fast, and when stuff like a background AV scan doesn't grind the whole laptop to a crawl.

I've already used a few 120GB 840 TLC drives when upgrading people's computers. So far performance seems very good and satisfaction is high, but I'll be sure to keep an eye on the situation.

Ditto for me too. Even most of my small business customers would be pushing it hard to get 1GB of data through their machines a day.

Power users are few and far between in the general scheme of things. And if you are a power user then you stump up the according cash to deal with it.

As for SSD performance I never take much heed of the raw MBps data. As long as its pushing 100MBps+ with near instant access times it will destroy the customers previous 120GB 5400rpm IDE Seagate from high orbit.:D

I think most people here don't realise the average HDD speeds most folks live with are around the 40-50MBps mark and that's being generous.
 
Ditto for me too. Even most of my small business customers would be pushing it hard to get 1GB of data through their machines a day.

Power users are few and far between in the general scheme of things. And if you are a power user then you stump up the according cash to deal with it.

As for SSD performance I never take much heed of the raw MBps data. As long as its pushing 100MBps+ with near instant access times it will destroy the customers previous 120GB 5400rpm IDE Seagate from high orbit.:D

I think most people here don't realise the average HDD speeds most folks live with are around the 40-50MBps mark and that's being generous.

Same here. Installed 3 of them so far in computer's friends and they are simply amazed with the real life performance. They are casual users and don't mind to benchmarks. The lifetime remains an uncertainty, only time will tell.

I think HardOCP should stick more to real life tests than canned benchmarks tests for SSDs. They have doing real life tests with GPUs that I appreciate. For SSDs, it's more like a review like any other website's review. My 2 cents.
 
I think HardOCP should stick more to real life tests than canned benchmarks tests for SSDs. They have doing real life tests with GPUs that I appreciate. For SSDs, it's more like a review like any other website's review. My 2 cents.

We appreciate your feedback, and we are working on several new concepts for our test regimen.
However, there are no other sites that provide steady state testing in the manner that we do, and we also feature a different perspective when testing with synthetics, not focusing on ridiculous IOPS numbers as much as the lower QD performance that matters.
Part of the reason for using some standard benchmarks is to provide users with repeatable tests that they can use as an example for their own gear. Proprietary traces can lead to many issues, and do not function with a filesystem present.
 
We appreciate your feedback, and we are working on several new concepts for our test regimen.
However, there are no other sites that provide steady state testing in the manner that we do, and we also feature a different perspective when testing with synthetics, not focusing on ridiculous IOPS numbers as much as the lower QD performance that matters.
Part of the reason for using some standard benchmarks is to provide users with repeatable tests that they can use as an example for their own gear. Proprietary traces can lead to many issues, and do not function with a filesystem present.


Please keep with your current methodology as it certainly opened my eyes (I own a 256Gb 830 Samsung as my C drive and a 128Gb 840 non Pro as my cache drive for my mechanical drives)
 
I dont quite understand how you come to the conclusion that the 120gb version vill fail after 3,5-4 years of light use?

120gb * 1000p/e cycles = 120 000gb total write endurance (actually a bit more with the extra space reserved for wear leveling...)

120 000gb/10gb a day = 12 000 days /365 = 32 years.

That would mean that you calculate with a write amplification of 10 wich seems a bit excessive with light usage. I doubt you would get write amplification above 1,1 with light usage, and even if you want to be conservative anything above wa at 3 is highly unlikely with anything close to normal usage patterns.
 
We did calculate with a WA of 10.
Depending upon the amount of free space WA can vary, and with the relatively small size of the SSD we were assuming that it would be at a decent fill for its life.
Also, ECC and other data protection algorithms can create wear, and with the nature of TLC we feel that these routines will become extremely active in the later stages of the drive life. This is gong to end up putting a background 'stream' of wear on the SSD, and this isn't accounted for in the JEDEC specs. They are only looking at purely cell wear.
In the end this is very speculative, of course. I feel it is always better to be very conservative when it comes to trusting a device, or recommending others to do so.
In 4 years we could have a lynch mob at my front door...
The bad thing is that there will be no clear 'breaking point'. It will just consist of rapidly increasing loss of data. A few unrecoverable bits here, a bit there, and we are looking at OS re-installs, corrupted images/documents/data, etc.
Users will likely not be suspicious of this, either, until a few issues in, and that is scary.
 
Last edited:
We did calculate with a WA of 10.
Depending upon the amount of free space WA can vary, and with the relatively small size of the SSD we were assuming that it would be at a decent fill for its life.
Also, ECC and other data protection algorithms can create wear, and with the nature of TLC we feel that these routines will become extremely active in the later stages of the drive life. This is gong to end up putting a background 'stream' of wear on the SSD, and this isn't accounted for in the JEDEC specs. They are only looking at purely cell wear.
In the end this is very speculative, of course. I feel it is always better to be very conservative when it comes to trusting a device, or recommending others to do so.
In 4 years we could have a lynch mob at my front door...
The bad thing is that there will be no clear 'breaking point'. It will just consist of rapidly increasing loss of data. A few unrecoverable bits here, a bit there, and we are looking at OS re-installs, corrupted images/documents/data, etc.

Well, looking at the ssd endurance thread at xtremesystems, that doesnt seem to be the way a 840 dies, they had no uncorrectable errors until the drive died. Blocks were retiered before they caused errors. This is the last data from their test:

http://www.xtremesystems.org/forums...25nm-Vs-34nm&p=5163560&viewfull=1#post5163560

I definetly understand why you would want to be very conservative when giving an estimated lifetime, but I still think it can be a bit misleading when say 3,5-4 years and dont explain that its a very conservative number and a more likely lifetime would be 10+ years, especially with light usuage that wont cause as much wa as a more demanding workload. But its your site and you write what you think is best ;)
 
The thread at XS is spectacular, and full of good info, but it needs to be framed correctly.
The majority of those SSDs have died from component failures, and not NAND failures. Not one has fallen into a read-only state.
We have no failure analysis that can give us a distinct reason why these drives are dying.

Also, and most importantly, these drives are not being tested for data integrity.

When a company follows JEDEC standards to predict the lifetime of an SSD they test the data retention. An SSD has to retain data, without power, for a year. This is the true test. Anything less than a year of data retention without power is deemed unacceptable. This data loss is due to leakage.
These SSDs are being tested with continuous power, nonstop. These testers have had drives fail due to being without power for a few minutes, that up until that point were running just fine.
This is indicative of the flaw of testing in this manner. This is also buried deep in that thread, there were even several attempts of testing data retention without power, but proved too time consuming.
If you were to take those SSDs and place data on them, then leave them without power for a few weeks, test with a hash, I would bet money many would not retain the data entirely.
Data loss on SSDs due to NAND wear is not dramatic, and doesn't constitute an entire failure typically. It involves small losses of data which become increasingly common.
The 840 that was tested showed absolutely no signs of pre-failure, no failure whatsoever. This is indicative of some type of other failure, many center around power components on the PCB.


EDIT: If the data that is being written in this testing, without removing power, were to be read back and verified at intervals, there is sure to be a number of data errors long before these SSDs 'die'.
 
Last edited:
Earlier in this thread I posted that I didn't feel 10GB was an accurate estimate of typical light usage. Many disagreed, but there hasn't been much actual data posted on either side yet.

I just had a followup for the first computer that I installed an 120Gb 840 TLC drive into a month ago. This is the main computer that handles administrative data/financial data for a construction company and sees use every day.

Samsung Magician reported the total data written to the drive as "0.1TB" and in the brief period I was there I didn't have a chance to dig around for a more detailed GB figure, but if we round up and assume 199GB had been written to the drive, that only works out to 6.633GB per day. Note that the written data includes all of the data that was written while installing windows, etc which after only a month is going to throw off the averages considerably. That, combined with rounding up, it's pretty clear that this computer isn't seeing anywhere near 10GB usage per day, in fact, much less than half that.

I'm going to continue keeping an eye on the computers that I've used these drives in. I would like to see additional people posting more actual data, so that we can get away from the unsubstantiated theories and start discussing useful information.
 
A performance comparison with the previous generation series 830 128GB would have been helpful.
 
Earlier in this thread I posted that I didn't feel 10GB was an accurate estimate of typical light usage. Many disagreed, but there hasn't been much actual data posted on either side yet.

I just had a followup for the first computer that I installed an 120Gb 840 TLC drive into a month ago. This is the main computer that handles administrative data/financial data for a construction company and sees use every day.

Samsung Magician reported the total data written to the drive as "0.1TB" and in the brief period I was there I didn't have a chance to dig around for a more detailed GB figure, but if we round up and assume 199GB had been written to the drive, that only works out to 6.633GB per day. Note that the written data includes all of the data that was written while installing windows, etc which after only a month is going to throw off the averages considerably. That, combined with rounding up, it's pretty clear that this computer isn't seeing anywhere near 10GB usage per day, in fact, much less than half that.

I'm going to continue keeping an eye on the computers that I've used these drives in. I would like to see additional people posting more actual data, so that we can get away from the unsubstantiated theories and start discussing useful information.

Mine has been on for 38 days and it's already been written 1.38TB on it. How about that? That's 36 Gigabytes everyday and 91 Gigabytes just the last 7 days. :)
 
Last edited by a moderator:
Mine has been on for 38 days and it's already been written 1.38TB on it. How about that? That's 36 Gigabytes everyday and 91 Gigabytes just the last 7 days. :)

That data is not particularly useful without any context.

Are you using any particular app that is causing the majority of those writes?
Are you contending that those writes are coming just from normal system activity?
My post referenced "light usage", do you feel your usage fits that description?
How much of the overall written data took place during initial system setup?

All of the systems that I've actually installed these drives into, most of which I've followed up on since then, are indeed showing very low usage that will not be a problem for TLC longevity. None even close to 10GB per day, with the highest usage at about a third of that.
 
That data is not particularly useful without any context.

Are you using any particular app that is causing the majority of those writes?
Are you contending that those writes are coming just from normal system activity?
My post referenced "light usage", do you feel your usage fits that description?
How much of the overall written data took place during initial system setup?

All of the systems that I've actually installed these drives into, most of which I've followed up on since then, are indeed showing very low usage that will not be a problem for TLC longevity. None even close to 10GB per day, with the highest usage at about a third of that.

I am measuring it with Processexplorer using the I/O Write bytes column. The guilty part is Firefox and i don't even have alot of tabs open, just about a dozen at most I browse alot and use youtube some. My computer has been on for five hours since system boot and Firefox has written about 6 gigs.
 
@dj christian
Go back to Process Explorer (v15.3), elevate (File > Show details for all processes), right click the column headers and add Process Disk > Read Bytes and Write bytes. Then check View > Format I/O bytes columns. Keep Process Explorer open (Options > Hide when minimized) for totals.

I/O Write Bytes = The number of bytes written in input/output operations generated by the process, including file, network, and device I/Os.
 
Last edited:
Thanks for the review Kyle. I am buying these via [H] Amazon to start replacing all the drives in our laptop fleet.

 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Back
Top