continuation

Recommendation for a performance 256GB consumer SSD?

16 posts in this topic

I'm building a new mid-tower Linux PC. In the process of picking a 256GB SSD. Have read many of the reviews on this site. Vertex 4 and Samsung 840 Pro seem to have the best performance after preconditioning. However they are both 512GB in the benchmark and it's unfair to compare them to other models which are all 256GB. Which 256GB SSD would you recommend?

Share this post


Link to post
Share on other sites

I've got a Samsung 830 256GB, a Crucial M4 256GB, and a Crucial M500 240GB. I can see very little difference. The 830 is probably the more reliable, as Crucial seem to have a few firmware bloopers. That said, I have a number of Crucial M4 drives, and with firmware updates I've had no data loss or issues yet.

I have had horrible results from OCZ SSD drives, though.

Share this post


Link to post
Share on other sites

We like the 840 Pro, Vector, Intel 520, Corsair GTX...kind of depends on your budget and what the deals look like at any given time. There will also be a new option next week to consider.

Share this post


Link to post
Share on other sites

Out of everything mentioned I doubt you could tell the difference in real-world use, so don't stress about it too much.

Share this post


Link to post
Share on other sites

Out of everything mentioned I doubt you could tell the difference in real-world use, so don't stress about it too much.

So true, for average daily tasks I cannot really see any difference between one of my old Intel X-25M and my latest Samsung 840 Pro (my benchmarks are a very different story). Reliability should be the greatest concern and Samsung, Intel and Crucial have all been good to me.

I look forward to looking at the new option that Brian alluded to. :)

Share this post


Link to post
Share on other sites

At Newegg, the specs for the Samsung 840 Pro 256GB

are the same as the specs for the 512GB model.

We've been experimenting with 4 x 6G SSDs

on PCIe 1.0 and PCIe 2.0 chipsets:

the scaling difference attributable to the

faster upstream bandwidth is almost linear.

So, if you're into MAX HEADROOM,

you should see almost 2x READ speeds

with 2 x Samsung 840 Pro in RAID 0.

Here's what we measured with ATTO, Direct I/O ENABLED,

on a PCIe 2.0 chipset, 4 x Samsung 840 Pro 128GB in RAID 0:

4xSamsung.840.Pro.SSD.RR2720.P5Q.Deluxe.Direct.IO.2.jpg

1846 / 4 = 461.5 MB/second per RAID 0 member (i.e. pretty good scaling w/ Highpoint RocketRAID 2720SGL)

MRFS

Share this post


Link to post
Share on other sites

We've been experimenting with 4 x 6G SSDs

on PCIe 1.0 and PCIe 2.0 chipsets:

That's very interesting.

I'm assuming the numbers shown in your graph are for sequential IO.

What about random IO? Can you use this setup to turn a bunch of Samsung 840 Pro into a poor man's FusionIO? Can you actually get FusionIO type of random IO performance out of this?

Edited by continuation

Share this post


Link to post
Share on other sites

We didn't run a complicated battery of tests;

we only ran a few tests to confirm a key hypothesis

by comparing PCIe 1.0 and PCIe 2.0 chipsets.

Here's the result we got with CrystalDiskMark

on the PCIe 2.0 chipset:

4xSamsung.840.Pro.SSD.RR2720.P5Q.Deluxe.CDM.jpg

Here's the result we got with AS-SSD:

4xSamsung.840.Pro.SSD.RR2720.P5Q.Deluxe.AS-SSD.jpg

Even on an older PCIe 1.0 chipset, the "feel" is much more snappy

when the C: partition is hosted by 2 x SanDisk 120GB in RAID 0,

but SanDisk SSDs use SandForce controllers which only excel

with compressible raw data.

Happily, the latest Samsung 840 Pro does not particularly care

if the raw data are compressible or not.

And, thus far no one has been able to refute our prediction

that WRITE ENDURANCE is increased by a factor of 2x

when 2 x SSDs are members of a RAID 0 array,

and it is increased by a factor of 4x

when 4 x SSDs are members of a RAID 0 array.

On the down side, Samsung did have a 64GB model 840 Pro,

but when I tried to order 4 of them, Samsung had

already discontinued that 64GB version.

FYI: there are quite a few other vendors who offer

6G SSDs in 60GB to 64GB capacities (if you are on

a limited budget):

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100018245%20600038519%20600038463%20600414916%20600038477&IsNodeId=1&Description=64GB%20SSD&name=64GB&Order=BESTMATCH

Edited by MRFS

Share this post


Link to post
Share on other sites

Samsung 830 256GB i think it is more reliable on you.In Finland i seen many people using that SSD and i also seen some people use to destroy it.In part of Helsinki many people has the best hard drive but there is some asiakirjojen tuhoaminen or destroying document services that not having doubt on destroying different kind of hard drive and i am thinking what about deleting i can't really understand that services.

Share this post


Link to post
Share on other sites

If it's a normal PC (i.e. not writing multiple 10's of GB per day) I'd go for the Samsung 840 instead of the Pro. You won't notice the performance difference, but the price difference is significant. And if it lasts only 10 years rather than 20 years due to the TLC NAND.. imagine what the money saved now could buy you in 10 years!

And' date=' thus far no one has been able to refute our prediction that WRITE ENDURANCE is increased by a factor of 2x when 2 x SSDs are members of a RAID 0 array, and it is increased by a factor of 4x when 4 x SSDs are members of a RAID 0 array.[/quote']

I remember that wierd dicussion. I hope we still agree on "write endurance of 2x128 GB = 1x256 GB" and "write endurance of 4x128 GB = 1x512 GB"?

MrS

Share this post


Link to post
Share on other sites

> I remember that wierd dicussion.

> I hope we still agree on

> "write endurance of 2x128 GB = 1x256 GB" and

> "write endurance of 4x128 GB = 1x512 GB"?

The point is the truth of the matter,

not whether we agree or not. :)

"Two blind men will fall into the same hole."

As I recall, the main point was that

"write endurance of 2x128 GB = 2x128 GB" (2 times the endurance of a single 128GB SSD)

and

"write endurance of 4x128 GB = 4x128 GB" (4 times the endurance of a single 128GB SSD).

comparing apples to apples.

The main point of a RAID 0 array is the speed increase.

See the ATTO graph above, which measured READs at >1,800 MB/second

(ASUS P5Q Deluxe, RocketRAID 2720SGL, 4 x Samsung 840 Pro 128GB in RAID 0).

6G / 10 = 600 MBps = MAX HEADROOM for a single 6G SSD and

a more practical ceiling is around 560 MB/second (measured).

Lastly, the 840 Pro has a 5-year factory warranty;

the 840 has a 3-year factory warranty.

Computing cost (at Newegg today) per warranty year for the

256GB 840 Pro and 250GB 840, we get:

$240 / 5 = 48.00 / year

$180 / 3 = 60.00 / year

Hope this helps.

Share this post


Link to post
Share on other sites

Of course RAID 0 is faster, that's not the issue (no need to present STR benchies). But why I stepped into this discussion is that if you write:

And, thus far no one has been able to refute our prediction that WRITE ENDURANCE is increased by a factor of 2x when 2 x SSDs are members of a RAID 0 array, and it is increased by a factor of 4x when 4 x SSDs are members of a RAID 0 array.

you make it seem like the RAID somehow magically increases write endurance. However, the reason for that increased endurance is just buying more NAND, which you could have just as well gotten with one larger drive. To first approximation, the write endurance of the RAID will not be any higher or lower than a single drive of the same capacity as the array. Can we agree on this?

Whether you prefer a larger single drive or several smaller ones in RAID 0 is a completely different question and has to be answered by each one individually (higher STR, similar random performance at low queue depths, higher risk of loosing a drive, slightly higher price, slightly higher power consumption). Personally I prefer a single small caching drive for most scenarios ;)

MrS

Share this post


Link to post
Share on other sites

> you make it seem like the RAID somehow magically increases write endurance.

Those are your words, not mine.

Then, in the next sentence you agree:

"the reason for that increased endurance"

So, it's not "magic": in a 2-member RAID 0 array,

each member performs only 50% of the WRITE load

that a JBOD drive must perform.

In a 4-member RAID 0 array,

each member performs only 25% of the WRITE load

that a JBOD drive must perform.

The comparison I am making is 2 @ 128GB with 1 @ 128GB

(NOT with 1 @ 256GB).

If you insist on comparing ONLY 2 @ 128GB with 1 @ 256GB,

go right ahead, but please be clear about that bias.

We have a PCIe 2.0 workstation with 16GB of SDRAM:

12GB of that SDRAM is a ramdisk using RamDisk Plus.

Because we load a 10GB database into that ramdisk

at every startup, and because we save that ramdisk

at every shutdown, the speed of those READs and WRITEs

is a consideration that is very important -- for us.

Using a single rotating HDD for this task

is totally out of the question.

This is particularly the case whenever system

software updates require a RESTART, which

then saves the entire database and then

restores the entire database BOTH at the

same time, one after the other,

before that computer returns to the Windows desktop.

The best we could achieve with 4 x 15,000 rpm

Hitachi 2.5" SAS HDDs was about 600MB/second.

As the raw data show (see above), 4 x Samsung 840 Pro

are doing READS in excess of 1,800 MB/second:

that's THREE TIMES faster than the Hitachi RAID 0

(exact number is 1,846.730 MB/second, ATTO benchmark).

For some reason that may be indicative of

false and/or misrepresentative advertising,

the SSD industry has been quite reluctant

to specify clearly the P/E cycles that

their various Nand Flash chips can handle.

If one does a little sampling of advertised

specifications at Newegg's SSD offerings,

for example, I believe you will find that

the number of P/E cycles is NOT stated

for the vast majority of those SSDs

being sold at a popular Internet retailer.

And, as I recall, user experimentation

was required to demonstrate that certain

SandForce controllers were designed to

inflict a severe reduction in performance,

if the Media Wear Index reached zero

before the factory warranty period expired.

Finally, it is generally being reported,

in technical reviews on the Internet,

that cost considerations are a major

motivation driving manufacturers to

utilize smaller and smaller die sizes

e.g. 34nm -> 25nm -> 20nm and so on.

Each shrink is reportedly accompanied

by a severe reduction in P/E cycles

e.g. 5,000 to 3,000 and still falling

some as low as 1,000.

Just do a little reading concerning

Triple Level Cell ("TLC") Nand.

In light of all such considerations,

when a computer needs very fast storage

that is NOT likely to "wear out" during

the factory warranty period, it has been

our approach to configure a RAID 0 array

of 4 x SSDs that have a 5-year factory warranty.

No magic: just common sense, in light of

the known properties of these Nand Flash

storage devices.

On the other hand, if one is paying the bills

with the advertising revenues of SSD manufacturers,

I can understand why one would want to avoid

offending them with any of the negative realities

of Nand Flash memory technologies.

Plug-and-Play is also suffering a direct hit,

but that's another story for another day

(HINT: Google TRIM RAID SSD)

Share this post


Link to post
Share on other sites

Here's an experiment that everyone can do, in less than 15 minutes:

Make a spreadsheet with the following columns:

Manufacturer / Model / Seq. READ Speed / Seq. WRITE Speed / Warranty Period

Refine a Newegg search to limit the results to

60GB and 64GB SATA III 2.5" SSDs.

How many manufacturers allow Newegg to disclose

the "Seq. WRITE Speed" and how many do not?

Crucial allows Newegg to report ~95MB/second,

but that spec is NOT reported for Corsair's models.

Share this post


Link to post
Share on other sites

Alright, let's try to bury the hatchet. There's obviously been some misunderstanding, since we agree on the obvious facts.

Had you written something like "Buying more drives for a RAID 0 gives you a speed increase and an endurance bonus due to the increased NAND capacity" I think there would not have been any discussion, because this is plain obvious. Whether this RAID is a good idea and the added endurance neccessary (or even beneficial) is a totally different question and depends on the intended application and the expected write load.

However, the way your original statement was written led me to believe that you meant going from e.g. 1x256 to 2x128 would increase endurance. Now I know that this is not what you meant (and is actually not what you wrote - my bad). But I still felt that others might misunderstand that statement too and make poor decisions based on it. That's why I wanted you to clarify what's being compared. And I think I've been clearly stating what I compared. Which is (IMO) not a bias, just starting from the point "I have x to spend on an SSD" or "I want/need y capcity of SSD storage". I think this is a pretty normal case (see e.g. thread title) and here it's usually not an option to just double or quadrouple the number of SSDs and hence cost. In your application example the prerequisites were different, of course.

Regarding your remaining statements: I'm well aware of P/E cycles, TLC etc. That's why I wrote "If it's a normal PC (i.e. not writing multiple 10's of GB per day)" when recommending the 840 over the 840 Pro. If you haven't you might want to check out Anand measuring the write endurance of the 840.

MrS

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now