Samsung 840 Series TLC 250GB SSD Review @ [H]

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,532
Samsung 840 Series TLC 250GB SSD Review - Samsung leads the way by introducing the world's first SSD equipped with TLC NAND into the market. The Samsung 250GB 840 Series SSD looks to shake up the market by providing an excellent dollar-to-performance ratio by leveraging low cost NAND and the high-powered MDX controller. Does its performance equal the value expectation?
 
Feeling kind of owned here, based off other reviews I bought a bunch of these. Now contemplating trying to return all of them........
 
Once again Kyle another great in depth review.

I jumped the gun last week and bought a 500GB 840 before waiting for your review to post. I have only had it installed for a week as my data / game / steam drive but so far it performs great.
 
Funny thing is I bought one of these a few days ago too. The guy at MC said we also have an 830 which is supposed to be faster. I discounted his opinion as sales people in retail rarely know much about what they are selling and stuck with the 840. I installed it and threw away the box and now wish I bought the pro or the 830. Though my 840 appears to be very fast. Granted I'm upgrading from an Athlon xp 2400 to an i5 3570 so this computer is lightning fast. I guess I'll need to buy another one in 5 years as my comp is not used much.
 
Funny thing is I bought one of these a few days ago too. The guy at MC said we also have an 830 which is supposed to be faster. I discounted his opinion as sales people in retail rarely know much about what they are selling and stuck with the 840. I installed it and threw away the box and now wish I bought the pro or the 830. Though my 840 appears to be very fast. Granted I'm upgrading from an Athlon xp 2400 to an i5 3570 so this computer is lightning fast. I guess I'll need to buy another one in 5 years as my comp is not used much.

Most salespeople at my local MC know their stuff and have interests in hardware. I do shy away from brand suggestions though as some are a little biased.
 
I looked at some of the benchmarks and it doesn't seem that much out of line when compared to that of the 840 Pro. I see some difference but it doesn't seem that bad, 10 to 15 MB/s less in some of the tests. Not bad if you ask me.

Personally speaking, the only reason why I bought an SSD is that it's faster than an old spinning type HDD. Sure, this SSD is slower than some of the more expensive SSDs that utilize MLC NAND but question that I ask myself when I look at SSDs is this simple question... Is it faster than a traditional HDD? If it is, then it's a winner in my mind.

All I care about really is... Will this make my machine boot faster? Apps start faster? If the answer is yes, then the SSD wins.

My comparisons are based upon an SSD versus an HDD.

I have an Intel 520 Series 250 GB SSD but I would like to get an SSD for a lesser used notebook that I have that would benefit from an SSD. Maybe getting one of these would be a good idea as versus one of the more expensive 840 Pro series SSD or even one of the Intel 520 Series SSDs.
 
I looked at some of the benchmarks and it doesn't seem that much out of line when compared to that of the 840 Pro. I see some difference but it doesn't seem that bad, 10 to 15 MB/s less in some of the tests. Not bad if you ask me.

Personally speaking, the only reason why I bought an SSD is that it's faster than an old spinning type HDD. Sure, this SSD is slower than some of the more expensive SSDs that utilize MLC NAND but question that I ask myself when I look at SSDs is this simple question... Is it faster than a traditional HDD? If it is, then it's a winner in my mind.

All I care about really is... Will this make my machine boot faster? Apps start faster? If the answer is yes, then the SSD wins.

My comparisons are based upon an SSD versus an HDD.

I have an Intel 520 Series 250 GB SSD but I would like to get an SSD for a lesser used notebook that I have that would benefit from an SSD. Maybe getting one of these would be a good idea as versus one of the more expensive 840 Pro series SSD or even one of the Intel 520 Series SSDs.

You hit the nail on the head friend. But forget about it being faster your HDD, look at this graph. http://images.anandtech.com/graphs/graph6428/51889.png

In a standard read-centric workload like the ones that average people like you and me will use the regular 840 is still one of the fastest drives to ever hit the AnandTech test bench. Hope you enjoy your SSD. :)
 
Actually the proprietary traces that Anand tests with operate outside of a filesystem, issuing Raw I/O to the device. This means that the testing also does not benefit from TRIM.
All of these recordings also do not place the SSD into steady state, so they are simply fresh out of the box results after a secure erase.
 
So these would be good ssd drives for games then? Contemplating on getting a 128 pro for os and programs and using my newly bought 840 for games? I could've saved money just buying a pro 256 but oh well..... :(
 
Actually the proprietary traces that Anand tests with operate outside of a filesystem, issuing Raw I/O to the device. This means that the testing also does not benefit from TRIM.
All of these recordings also do not place the SSD into steady state, so they are simply fresh out of the box results after a secure erase.

Thanks for the comment Hugh. So I think the point you are making is this: AnandTech's storage bench is not a realistic test for average people like me who are using Windows 7 with TRIM. Is that right?

If you had to pick one or two benchmarks in your [H] testing suite, which one would you choose? I want those benchmarks to be a rule of thumb for how good an SSD is compared to other SSDs.
 
I'm really surprised to see the 840 series (even 840 Pro) do so well compared to the Corsair Neutron GTX. I'll be considering sending back my Neutron GTX and going for the 840 Pro unless I'm missing something
 
Hate to nitpick but the NAND isn't Toshiba as referenced a few times throughout the article.

That aside I got the 250GB from Newegg for $150 a few days ago and am very satisfied with it. Yeah, the writes could be better but the reads and pricing make up for it IMO.
 
This standard 840 with TLC, I just do not trust it in the long term, specially for heavy use.

The m4 is a good alternative.
The 840 Pro is still too expensive, and hardly any faster than the m4 in practice.

NAND cell's function as capacitors the more electrons you for a certain value in a cell stops the more reliable it is.

1 bit SLC has two statuses on or off
2 bit MLC has four statuses on, two intermediate positions or off
4 bit TLC has eight statuses on six between positions or off

A SLC cell has 100 electrons for 1 bit.
Ware MLC you have 25 electrons for 2 bits
But whit TLC you only have 12.5 electrons for 4 bits

TLC for me is something I for one have no confidence in, for the long term, and with one or two process that shrinks the number of electrons decrease even more, TLC is something that might not even be reliable / possible to use.
Read here LSI talks about future flash and its problems.

Maybe the problem is in practice not so bad, especially for a HTPC or other light use, your properly save, as long as he does not get used for things like Torrent downloads, but I would not risk it, and i just buy a m4 instead!
 
Since I have an X58, would going to a more expensive faster drive have made a difference? The answer is none at all.

Simply put: This drive is a great value for upgrading an older system, or anything with less than 6GB SATA.

I bought the 250GB version for my boot drive. It max'd out my controllers bandwidth on both read and write (250-275MB/S).

When I decide to build a new system with the latest and greatest speed hardware, this will become a storage drive but until then, I would have wasted money buying something 'better'.
 
I looked at some of the benchmarks and it doesn't seem that much out of line when compared to that of the 840 Pro. I see some difference but it doesn't seem that bad, 10 to 15 MB/s less in some of the tests. Not bad if you ask me.

Personally speaking, the only reason why I bought an SSD is that it's faster than an old spinning type HDD. Sure, this SSD is slower than some of the more expensive SSDs that utilize MLC NAND but question that I ask myself when I look at SSDs is this simple question... Is it faster than a traditional HDD? If it is, then it's a winner in my mind.

All I care about really is... Will this make my machine boot faster? Apps start faster? If the answer is yes, then the SSD wins.

My comparisons are based upon an SSD versus an HDD.

I have an Intel 520 Series 250 GB SSD but I would like to get an SSD for a lesser used notebook that I have that would benefit from an SSD. Maybe getting one of these would be a good idea as versus one of the more expensive 840 Pro series SSD or even one of the Intel 520 Series SSDs.

To me the goal of SSDs is to offer RELIABLE storage that's faster than legacy drives, reliability is my number one concern, if it's not met then I wouldn't even consider the second feature, so my concern is more about the use of those TLC chips.

Considering the 840Pro itself probably takes an endurance hit from the shrink to 21nm vs 830 line's 27nm process, adding TLC to the mix - stuff that they put in cheap keychains - makes me a bit uneasy, I'm not willing to be the guinea pig for the product.

I'd be glad to see these fears proven wrong by early adopters that may still have working drives 10yrs from now after constant use, we'll see.
 
I think a web bug exists on page 2 of your 840 Pro review. Under features I see the 840 chart rather than 840 Pro. This bug is with the Pro review.

www hardocp com /article/2012/12/10/samsung_840_pro_ssd_review/2

I love the new 840 review. Thank you.
 
If it lasts three years, I'm happy. By the time three years comes about I'll probably be in the market to get another drive that's bigger (more storage capacity) and cheaper. I'll bet that most of you guys are going to be doing the same in three years, that is, buying a new SSD for the latest and greatest.
 
I'd be glad to see these fears proven wrong by early adopters that may still have working drives 10yrs from now after constant use, we'll see.

LOL. These statements kill me. Technology will far eclipse/obsolete the life of NAND for 99% of users in this market. Are you still using old-ass 10yr old IDE drives in your system? Do you have critical data on those IDE drives from 10 years ago? If so,I can't help you then.
Far, far too much emphasis is put on reliability projections a decade from now. It's doubtful SATA3 will be offered on a motherboard ten years from now. And WTH would you want it offered? They'll have far better technology out. Keep living in fear. I've been running/enjoying SSDs for 5 years now.
 
I have the 500GB model but haven't installed it yet. It's not like I'll hammer the drive with random 4KB writes after the drive is full like these reviews do.
 
LOL. These statements kill me. Technology will far eclipse/obsolete the life of NAND for 99% of users in this market. Are you still using old-ass 10yr old IDE drives in your system? Do you have critical data on those IDE drives from 10 years ago? If so,I can't help you then.
Far, far too much emphasis is put on reliability projections a decade from now. It's doubtful SATA3 will be offered on a motherboard ten years from now. And WTH would you want it offered? They'll have far better technology out. Keep living in fear. I've been running/enjoying SSDs for 5 years now.

As a matter of fact I am still using my old 80gb hard drive from pre-2000 as my main drive, with a couple slow green drives for storage/backup.

Setup works fine, no reason to replace it, but an SSD drive would make a great upgrade for it. That said, the safety of my data is more important than the performance gain so the replacement would need to be reliable, I hate the hassle/worry of re-installs and backups.

Not everyone likes to spend their money on cutting edge stuff, especially luxury products like what an SSD drive is currently. But to each his own, different needs.
 
Thanks for the comment Hugh. So I think the point you are making is this: AnandTech's storage bench is not a realistic test for average people like me who are using Windows 7 with TRIM. Is that right?

If you had to pick one or two benchmarks in your [H] testing suite, which one would you choose? I want those benchmarks to be a rule of thumb for how good an SSD is compared to other SSDs.

Yes, not to be disrespectful to Anand but here is how they do it. They record a period of usage in a typical filesystem environment (NTFS) with the ipeak utility. Then they replay this trace without a filesystem, this is called 'Raw I/O'. The behavior of an SSD with and without a filesystem is very different, especially with the applications. Applications use a file system. So you have issues for a number of reasons.
1. Without a filesystem there is no TRIM command generated. The filesystem tells the physical device which files are erased so that they can be cleaned up before re-write.
2. The internal mechanisms of the SSD have several processes that run during idle/quasi-idle times. These are GC and other housekeeping routines (static data rotation, ECC, etc). Recording a two week long trace and replaying it very quickly (the 'long' version of their test takes 8 minutes) does not leave time for these periods of rest for correct SSD function.
3. There is also the concern that if you record a QD1 environment for two weeks and play it back in a few minutes, there will be a ridiculously high amount of QD to 'simulate use'. SSDs internal functions operate differently from controller to controller, and also with different NAND flavors.
4. They are not testing in steady state conditions. They are only testing FOB (fresh out of box) results. The test does not even fill up the device once. Therefore no internal clean up algorithms can be triggered.
5. The ipeak application doesn't even use the same data as the application recorded. For instance, if I were to record a Photoshop session with some incompressible picture files with the utility and then play it back, the utility does not use the pictures that I was editing. It does not even use the files that were used for the application or the OS during that time. The utility creates its own data that does not resemble the real data recorded. This filler data is highly compressible, thus controllers that rely upon compression get an unfair boost. They are getting fed compressible data no matter what.

So, no TRIM, no filesystem, no internal SSD functions tested. No steady state. Filler data used instead of actual application data. Unrealistically high QD when the trace is played back in fast-forward. This is why we do not use a similar application with our benchmarks.

We test in steady state with a filesystem, with TRIM and the internal functions triggered by steady state, in real-time and with the actual applications. We feel that we give you, the readers, an actual correct representation of results that you can, and will, experience in every day use.

I cannot pick one test. I use all to formulate an opinion, as each highlights various aspects of performance that are relevant to the end user experience.
 
Last edited:
Even if the tests aren't perfect, using the same test across several drives is still very valuable information. My favorite is are Anand's latency plots ... I'd rather have a drive with lower throughput and consistent latency, than a drive that has 50% more average throughput but has latency hiccups.
 
Status: wantondstrction is offline
Hate to nitpick but the NAND isn't Toshiba as referenced a few times throughout the article.

Jesus, it is Samsung Toggle I was just typing away when I put Toshiba Toggle, I will get this corrected. Long nights and early mornings when I am writing :) Thanks!

Technology will far eclipse/obsolete the life of NAND for 99% of users in this market. Are you still using old-ass 10yr old IDE drives in your system? Do you have critical data on those IDE drives from 10 years ago? If so,I can't help you then.

Good point.

t. It's not like I'll hammer the drive with random 4KB writes after the drive is full like these reviews do.

We do not fill the SSD entirely. We fill to 75% per SNIA specifications for consumer SSD testing. There is a huge difference between full-span write testing (100% fill) and 75% fill.

Simply put: This drive is a great value for upgrading an older system, or anything with less than 6GB SATA.

Actually the benefit is more from latency that throughput. So yes, there is a benefit with faster drives with better latency. SSDs that have better overall latency provide enhanced random read and write speed, which is the majority of your typical usage pattern (80-90%). These speeds rarely go over 40MB/s, which is nowhere near saturating the SATAII bus.
This SSD is still better than many SSDs out there, and when it comes to HDD it is miles better.
 
Even if the tests aren't perfect, using the same test across several drives is still very valuable information

I do not see value in running tests that are totally unrealistic, even if you test a million drives with it.

Would I test a Yugo on a motocross course with 50 foot jumps, then test every other subcompact vehicle on the market on the same course and declare that I can tell how they handle in the city? Of course not.

I'd rather have a drive with lower throughput and consistent latency, than a drive that has 50% more average throughput but has latency hiccups.

The consistency testing that they are conducting uses FIO, which also operates outside of a filesystem. No TRIM. They are also testing with 100% full span writes, which reduces the ability for the SSD to run GC. This is enterprise testing on consumer SSDs.
 
Last edited:
The consistency testing that they are conducting uses FIO, which also operates outside of a filesystem. No TRIM.

"They"? Who is using fio?

And fio can be used in raw mode or filesystem mode. And even in raw mode, it is possible to manually TRIM LBA ranges on a device, using an external program.
 
Last edited:
ATTO runs independent of the file system which allows it to operate outside of any file system restrictions or overhead.

This is incorrect. ATTO creates a test file (benchtst.$$$) of size "Total Length" (default 256MiB) in the filesystem on the specified drive and uses that for its tests. This is easy to verify by just watching the file appear in a file browser window while ATTO is testing (it disappears as soon as the test completes).
 
Last edited:
"They"? Who is using fio?

And fio, like IOMeter, can be used in raw mode or filesystem mode. And even in raw mode, it is possible to manually TRIM LBA ranges on a device, using an external program.

"they" is Anand. Yes you can manually TRIM the SSD, but during typical SSD usage there is a constant stream of TRIM commands, users do not have to manually send the TRIM command at intervals.
 
You might wish to check their comments in their articles. They discuss results with a guy who inspired some of their testing with FIO. Still wanna bet? :)
 
You might wish to check their comments in their articles. They discuss results with a guy who inspired some of their testing with FIO. Still wanna bet? :)

Yes, I still want to bet. You can ask Kristian (Hellhammer) on the anandtech forums, unless you happen to have Anand's contact info.
 
Last edited:
I do not have Anands direct contact, but you can feel free to ask.
I can dig through and find the comments where they discuss it, but i will allow you to do the legwork :)
 
To me the goal of SSDs is to offer RELIABLE storage that's faster than legacy drives, reliability is my number one concern, if it's not met then I wouldn't even consider the second feature, so my concern is more about the use of those TLC chips.

Considering the 840Pro itself probably takes an endurance hit from the shrink to 21nm vs 830 line's 27nm process, adding TLC to the mix - stuff that they put in cheap keychains - makes me a bit uneasy, I'm not willing to be the guinea pig for the product.

I'd be glad to see these fears proven wrong by early adopters that may still have working drives 10yrs from now after constant use, we'll see.

10 years? Seriously? I don't think I've used any regular HDD for even 5 years before replacing them for something faster/larger. Sure, I've got a 6.4GB drive still spinning and other oddities, but I don't see why you would exaggerate the necessary/useful lifespan of an OS drive so much.

The whole emphasis on reliability seems a bit overboard, and don't get me wrong, I'm the sort of person that still wouldn't recommend Sandforce drive due to all the firmware shenanigans they went thru but come on... These drives are aimed at light users or laptops once they start building them in a stick form factor like the 830... Users are gonna keep their data elsewhere and hopefully they'll do a backup of their OS once on a while, I don't see what the big deal is if the useful lifetime of the drive turns out to be 5 years instead of 10, it'll probably still outlast the type of system it's going into.

Demanding users should obviously know their needs and do their research and opt for higher end MLC drives but this view of the 840 like it's some sort of evil consumer trap seems very misguided, if anything it's gonna help everyone as it further pushes NAND pricing down.
 
Last edited:
As a matter of fact I am still using my old 80gb hard drive from pre-2000 as my main drive, with a couple slow green drives for storage/backup.

Setup works fine, no reason to replace it, but an SSD drive would make a great upgrade for it. That said, the safety of my data is more important than the performance gain so the replacement would need to be reliable, I hate the hassle/worry of re-installs and backups.

Not everyone likes to spend their money on cutting edge stuff, especially luxury products like what an SSD drive is currently. But to each his own, different needs.

Wait wait wait, the hassle of backups? So you're actually banking on the reliability of a ten year old drive to keep your data safe because backups are hard? That may be the funniest thing I've read all week. Anything that you aren't backing up is simply infinitely more at risk regardless of what kind of drive it's sitting on. All drives fail, it's not a matter of if but when. If you haven't had several drives fail you then you're simply beating the odds, and highly at risk.

If we're simply talking about reliability in terms compatibility or uptime, Samsung's controllers have proven their worth regarding the former and flash health isn't hard to track thru SMART attributes... If the drive's gonna fail due to eroded flash it's still gonna be far more predictable and salvageable than a mechanical drive which may fail catastrophically at any given time, moving parts tend to do that.
 
Last edited:
I do not see value in running tests that are totally unrealistic, even if you test a million drives with it.

That's like saying no testing on the 'net matches my exact usage pattern, so they are all totally unrealistic for me. At some point even when the testing is synthetic, conclusions can be drawn.
 
I think my 840 Pro will last me til the next version of SATA comes out :D
 
Back
Top