While just about any
SSD will make your average computing experience a fair bit more awesome, it takes a really unique device to make said experience Animal-Style-Triple-From-In-N-Out special. Up until now, the general consensus was that Intel's
X-25M was the crème de la crème, but it seems that Corsair's recently launched
P256 may just be giving that very drive a real run for its money. After seeing a pre-production unit
deliver some respectable early results, we figured it prudent to pass along
bit-tech's full-on review. In most cases, the 256GB P256 either topped or fell just behind Intel's 80GB unit, though the drive did seem to suffer a bit in the random read / write tests. Still, critics felt comfortable recommending the drive, and while pricey, noted that it offered better value per gigabyte in comparison to similar 256GB units on the market.
posted Jun 6th 2009 8:08AM
"Animal-Style-Triple-From-In-N-Out"
Eh... What?
this is the new 'first'.
supposedly it's a LA thang. Don't you watch Conan O'Brien?
Oh it's just a fat ass burger for eating.
It's "Animal-style-triple-DOUBLE-from-In-N-Out". C'mon Engadget.
Love the In-N-Out reference, makes me miss California
Mmmmmm. So happy to live here :x Especially if I actually feel like having a burger.
Excuse me, Darren, but I believe that would be a 3-by-3 Animal Style.
animal style 3x3 .... california ftw
With animal style fries?
For those on the East Coast, In-N-Out is a burger chain in California, Nevada, Arizona, and Utah. They have a very small menu and serve great burgers. They're also one of the very few remaining family-owned chains left.
'suffer a bit in the random read / write tests'.
so its crap. :P
Exactly. The random writes are what make the damn things stutter. Check out the graph the article has here:
http://www.bit-tech.net/hardware/storage/2009/06/05/corsair-p256-256gb-ssd-review/7
4.5 megabytes per second versus the intel's 39. It's no contest. Yes, this drive is maybe 3rd place (4th if you count the X25-E), but it's a distant third place.
Might want to click the read link. While it's nowhere near as good as the Intel drive, it beats the VelociRaptor in random writes. At the very least it won't have the stuttering problems of previous SSDs.
No, this drive uses the new Samsung controller like the OCZ Summit series, and both have the best random read and write outside of the Intel X25M, and are far cheaper. Also, while the X25M has great random write performance, it is 2-3X slower in sequential write than other drives.
To reinforce, no, definitely not crap.
So no drive has it all, when a drive performs well on all scales mentioned then I will be satisfied.
@ mirakutea
You hit the nail on the head man, the MFRS have to compromise in either sequential, random or both depending on their priorities. It's only a matter of months before drives start showing up that have it all, plus, all the time we wait, the cheaper the tech gets too.
But there will still be manufacturers that optimize in one direction or the other, compromising in the other, making the drives that have it all look like less than they are. There are drives that have it all now, they just can't match certain drives that compromise for a single metric, in that single metric.
Corsair makes excellent products.
'Nuff said.
Repeating comment beolow:
Random write speeds:
Intel X-25M: 39.49MB/s
Corsair P256: 4.67MB/s
Random Write Latency:
Intel X-25M: 0.30ms
Corsair P256: 2.5ms
Corsair now also makes failure products :P.
Dont listen to the idiot above me, the situation is far more complicated than he suggests.
The Corsair P256 and OCZ Summit both use the new Samsung controller, and while they don't have the random write performance of the Intel X25M (overkill), their random write performance is 1000X better than SSDs. that use the old JMicron controller, and they don't experience any of the write stuttering whatsoever.
Also, as many review sites have covered, the random write performance of drives using the new Samsung controller or an Indilinx controller are still blazing fast, and for non-server use Intel's X25M random write performance is total OVERKILL and doesn't provide any TANGIBLE BENEFITS over those two other controllers for desktop use. As long as the SSD isn't using JMicron's old controller (which literally has random write performance on the order of 1000X worse than the Indilinx), you won't notice a difference.
In addition, Intel's over-optimizing for random write performance led the X25M to have a sequential write speed of only 75MB/sec --- whereas the Indilinx or new Samsung controller-based drives doing over 200MB/sec (while still having plenty of random write performance)
==== Lastly, my recommendation for EVERYONE is to read Anandtech's 20+ page SSD article in addition to Anandtech's, PCperspective's, and Bit-tech's reviews of SSDs. They provide a ton of very helpful information to understand the complexities of SSDs ===
@loosely_coupled - Thanks for the break down, very informative. Its nice to see actual information in a comment rather than typical flame bs.
@loosely_coupled
Thanks for that post, helped clear some confusion up for me.
And just from personal experience, I have always been satisfied with Corsair products to date.
Since when was "bit-tech.net" called "PC Perspective"?
I had one of the first reviews of this drive running on a Mac Pro - but I am a small IT dude....who would pick that up. :-)
http://www.2s2d.net/2009/05/15/corsair-p256-cmfssd-256gbg2d-on-a-mac-pro/
Check it out.....
Random reads/writes matter.
Indeed they do.....
Not sure wtf they're talking aboot, but consider me converted.
Bah, its MLC. Those things are so unreliable you might as well buy one of those old hard drive thingies. Sure they are a nice way of getting SSD's into the mainstream at a somewhat competitive price but people will run a mile when they all start to fail in a couple of years. MLC is only good for about 10,000 erase cycles while SLC is 10-100 times that and also much faster. SLC is more expensive but imho definitely worth spending the extra money on a superior product that will last
So it's a big fat failure?
With good wear leveling these MLC drives will last 10 years or so and it's only going to improve as the technology matures, SLC is great and all but it's 3 times as expensive and not that much faster.
Quite spreading disinformation. MLC SSDs are not "un-reliable" and have a far longer MTBF than any HDD. On average use, most sources say newer MLC SSDs with good wear leveling algorithms should easily last 10 years. And even when the time comes, the drives just convert into read-only.
To those talking about wear leveling,
Can you get wear leveling to work with a raid setup? I was always little confused about that. At some point I thought I read that it wasn't possible, or that raid screws up some types of SSD maintenance, but stuff keeps changing. Anyone know?
Repeating the remarks above. MLC is the future of SSDs for the moment. SLC is going to be limited to very specific markets where the density doesn't matter, the price is irrelevant and every bit of performance is critical. You can read this as "not desktops or laptops for normal people". Doesn't mean this will always be true, but in the near future if you're going to buy an SSD, its going to be MLC. And yes, while the number of erase/write cycles an MLC block can sustain is lower than SLC, its entirely high enough. If you don't believe me, read up on the subject elsewhere. Start with Anandtech.
Anyway, sure SSDs can be used in RAID. The RAID controller will only deal with the "logical" location of the data. The wear-leveling occurs BELOW that, and is invisible to the RAID controller.
If I use my Samsung P256 for 10 years it will be the first storage device I've ever used for more than 4 years...
Truly, I've been using my P256 for 6 weeks of intensive graphic design and photography (Lightroom + 50MB DNG & NEF files) and no other HDD even comes close! I would have gotten an Intel device if the capacities were >160 GB but I needed the 256. I am overwhelmingly impressed. The drive upgrade beats a 4-5x improvement in processor speed for my workflow. You can't buy that performance increase AT ANY PRICE (except a SSD) today.
I cannot stress enough how awesome SSD computing on a MacBook Pro is. (Yes, you will have to do a careful teardown of your machine with the right patience and tools to do it right, but you're all hackers, right?)
The drive is pretty good on its own...but for half the price I found a Velociraptor 300gb performed just as good if not better.
no it didnt.
not even close in real world usage.
did you find some special raptor that is way better than the one everyone else can buy?
I ran a Simple TEST with the Velociraptor 300gb and the Corsair and posted the results on my site to which the link is above. I dont make em...I just use them and ran this simple test I could find for a MAC....the results were impressive.
I'll say it again:
Random reads/writes matter.
Ryan - YES Random reads/writes MATTER very much. As I said...it was a simple test to do a read and write with files of different sizes...infact the test was for Video - unfortunately it was the only free utility I could find to do the test. I am sorry you are not happy that the review did not show random reads and writes.....but it did what it did. I am not trying to compete in reviews....just doing my own thing for my site.
But that review is worthless and misleading. If you're not testing SSD's main advantage for desktop use and then say it's not worth it, what kind of analysis is that? The fact is that in general use you're not writing 128MB files over and over, you're writing/reading hundreds of small files randomly. So your review basically means the following. If you like copying a blu ray rip between two hard discs over and over again SSDs are not worth it. If you like using your machine they may or may not be worth it because I was too lazy to do the correct tests.
Random write speeds:
Intel X-25M: 39.49MB/s
Corsair P256: 4.67MB/s
Random Write Latency:
Intel X-25M: 0.30ms
Corsair P256: 2.5ms
it's not even in the same order of magnitude.
and to say that it doesn't suffer from slowdown like the Intel drives do is bullshit.
It's a fundamental problem with flash based drives to slowdown once they have written to every block once.
That is why the TRIM command needs to be FULLY supported and the blocks need to be cleaned when they are eraased, not when they are being re-written.
this "review" has FAIL written all over it.
They didn't even mention the X25-E which I have as my boot disk (64GB).
Nevermind, I do see the X25-E, but it is missing from many of the graphs.
Numbers are worthless if you don't understand their meaning.
The Corsair P256 and OCZ Summit both use the new Samsung controller, and while they don't have the random write performance of the Intel X25M their random write performance is 1000X better than all the other SSDs that use the old JMicron controller, and they don't experience any of the write stuttering whatsoever.
As many review sites have covered, for non-server use Intel's X25M random write performance is total OVERKILL and doesn't provide any TANGIBLE BENEFITS over the new Samsung or Indilinx controller. As long as the SSD isn't using JMicron's old controller (which literally has random write performance 100X worse than any of these drives), the random write performance won't be an obstacle of any kind.
In addition, Intel's over-optimizing for random write performance led the X25M to have a sequential write speed of only 75MB/sec --- whereas the Indilinx or new Samsung controller-based drives doing over 200MB/sec (while still having plenty of random write performance)
@loosely_coupled
FAIL!
so it's overkill if you optimize for Random reads/writes which is the VAST MAJORITY of everything that a computer does that actually bottlenecks a drive (very little is stored sequentially on an SSD thanks to block level wear leveling which basically takes advantage of the ultra low "seek" times of flash memory and spreads blocks of even LARGE files all over the "disk") but okay to optimize for artificial benchmarks that do nothing?
oj, i see, you really DON'T know what you're talking about.
there are no sequential reads and writes of anything larger than a block size on an SSD drive.
The very same Anandtech article you cite will explain this very well.
for reference you can read starting here:
http://www.anandtech.com/storage/showdoc.aspx?i=3531&p=8
this will explain that, once again, you can't erase anything less than a whole block at a time, so in order to prevent the same block (the "first") from being the target of all the erases on a non-full drive, SSD controllers employ block level wear leveling which basically splits every file larger than glock size and sends it to an algorithmically selected block that hasn't been written to as much and thus creates the necessity for RANDOM read access being the MOST important metric for an SSD drive.
If random read weren't important, and only sequential read were important, than the very same JMicron controllers you are bagging on would be perfectly acceptable for use in desktop SSD drives, because they have very good sequential read performance even when "used"
citation here: (see drives #3&4)
go to page 14 as i can't enter enough url's to directly link
instead, we get the famous "stuttering" of JMicron drives due to their TERRIBLE random read performance, and we get this while just trying to open a simple application.
citation here: (look for the second table and read directly above and below it as well as the numbers on the table)
go to page 17 as i can't enter enough url's to directly link
you also say that the Intel drive has a sequential write speed of 75MB/s which is patently false as proved by this wonderful graph from the same article which shows it to be 191.7 MB/s. Or basically within 10% of the indilix/samsung drives.
citation here: go to page 24 as i can't enter enough url's to directly link
now go to look at random write speeds and see that the intel has a speed of 23.1 (now a bit higher evidently by the other tes but lets keep the same numbers from the article)
or even then nearly 500% higher than the corsair drive.
random matters
more so than ANY sequential number ever will.
sequential numbers are published (by anand's own admission) as "sexy big numbers" to sell drives.
That's it.
no thanks,
Sorry, but you're the one who doesn't know what they're talking about. When people talk about sequential vs. random performance they mean "sequential or random as far as the host is concerned". Regardless of the fact that the host doesn't know where the data is actually located because of wear leveling, they still run the tests that way and see differences. The reasons would have to do with the controller, and its ability to predict the locations of things, and the way the cache works, and the ability of the controller to parallelize the operations when they are sequential, etc.
Since when do you write to every block once?
And there are several safe and documented procedures that can restore Samsung/Crucial + Intel drives to new condition by booting off a cloned external drive, writing one volume-filling file to the SSD, erasing the file, reformatting the SSD and then re-cloning the drive & make it the boot device again. Depending on how much capacity you're truly using on the device you need to do this every 9-12 months. And the fact that the whole process takes about 60 minutes on my MacBook Pro.... well, it's faster than defragging a Windows (which you have to do every week!)
In my book the 100x increase in speed over a 7200.3K mechanical disk for desktop use is more than worth it.