Six-Drive SATA III SSD Round-Up Shows Big Gains

MojoKid writes "Solid state drives have gone from essentially non-existent on the desktop to the preferred storage medium of enthusiasts and workstation professionals in less than three years. Three of the drives featured in this six-drive SATA III SSD round-up consistently offered 'best-of-class' performance throughout testing, with speeds in excess of 500MB/s for read and write throughput. OCZ's Vertex 3 Max IOPS, Corsair's Force GT, and the Patriot Wildfire all feature the same SandForce SF-2281 controller and synchronous NAND flash memory. These drives offered the highest transfer rates in the majority of tests, though performance does drop off as the data gets more incompressible."

Comments

However, something important to keep in mind

[Sycraft-fu]

At this point, all SSDs are basically "fast enough" for desktop usage. You notice a major difference between an SSD and a HDD. You don't notice much, if any, difference between a lower and higher end SSD on the desktop.

The same is not true on servers, of course, the heavier random load makes IOPs a big deal in various servers (databases particularly).

While I'm certainly not saying don't get one, I'm saying don't dump your SATA II SSD if you have one for these, and don't pass up a SATA II SSD if it is on sale.

Re:However, something important to keep in mind

[Solandri]

At this point, all SSDs are basically "fast enough" for desktop usage. You notice a major difference between an SSD and a HDD. You don't notice much, if any, difference between a lower and higher end SSD on the desktop.

A large part of the reason SSDs are faster than HDDs is their low latency (has a big impact on small file read/writes). However there's another big reason you don't notice much difference between lower and higher end SSDs: We're using the wrong metric.

SSDs and HDDs benchmarks are almost universally given in MB/s. The problem is, people don't perceive speed in MB. They perceive it in seconds. The computing tasks you need to get done are almost never "I can wait 1 second. How much data can my computer crunch?" They're of the type "I need to crunch 1 GB of data. How many seconds will that take?" So the correct metric we should be using is s/MB.

But it's the same number! Why should this make a difference? Because when you invert a metric, the big numbers become small numbers, and the small numbers become big numbers. e.g. Say you have a HDD which can read 100 MB/s, a cheap SSD which can read 200 MB/s, and an expensive SSD which can read 500 MB/s. So in 1 second, the HDD reads 100 MB, the cSSD 200 MB, and eSSD 500 MB. Expressed in MB/s you gain 100 MB/s switching from HDD->cSSD, and a whopping 300 MB/s switching from cSSD->eSSD. Switching from cSSD->eSSD gives you 3x the benefit of switching from HDD->cSSD! So the extra money for the expensive SSD is definitely worth it! Right?

Hold on. Invert to s/MB and say you need to read 1 GB. The HDD takes 10 sec, the cSSD 4 sec, and the eSSD 2 sec. Switching from HDD->cSSD saves you 6 seconds. Switching from cSSD->eSSD only saves you 2 sec. So in terms of time you spend waiting, the HDD->cSSD switch saves you 3x as much time as the cSSD->eSSD switch. The vast majority of your time saved can actually be obtained from the switch to the cheaper SSD. The next step switching to the expensive SSD only gives you a marginal improvement. (Even if you insist on using relative measures of time, the cheap SSD still wins. 10 sec to 4 sec is a 60% reduction in time. 4 sec to 2 sec is only a 50% reduction in time.)

Anandtech basically stated as much in a recent SSD review. They admitted that in real world use (i.e. benchmarks measured in seconds), there really isn't much difference between the different SSDs. But reviews with benchmarks showing all products having nearly the same result doesn't get people coming back to read more reviews. So to hype people up, reviewers invert the scale and measure in MB/s to exaggerate small differences. Differences which for the vast majority of people are so small as to be nearly meaningless in their real-world computer use.

The same thing crops up with fuel mileage in cars. Fuel consumption is actually gallons per mile. But because the U.S. measures it in miles per gallon, it exaggerates the benefit of high mileage vehicles. If you ask a dozen people which saves more gas, switching from a 14 MPG SUV to a 25 MPG sedan, or switching from a 25 MPG sedan to a 50 MPG hybrid, I will bet nearly all of them will say switching to the hybrid saves more gas. After all, 50-25 = 25 MPG improvement, while 25-14 = only a 11 MPG improvement. But if you drive 100 miles:

14 MPG SUV = 7.1 gallons used
25 MPG sedan = 4 gallons used
50 MPG hybrid = 2 gallons used

Surprise. The 11 MPG improvement switching from the SUV to sedan saves you 3.1 gallons per 100 miles driven, while the 25 MPG improvement switching from sedan to hybrid only saves you 2 gallons. The metric we should be using is GPM, not MPG. The rest of the world measures fuel consumption in liters per 100 km for this reason. (A consequence of this is that if we as a nation wish to lower our fuel consumpt

Re:However, something important to keep in mind

[Anonymous Coward]

Surprise. The 11 MPG improvement switching from the SUV to sedan saves you 3.1 gallons per 100 miles driven, while the 25 MPG improvement switching from sedan to hybrid only saves you 2 gallons

But (gah!) you're just making the same error you're complaining about - looking at absolute changes rather than relative ones! Switching to the hybrid does save you more gas, when considered as a percentage of how much gas you were using before.

Maybe I'm just strange, but I've never considered such figures in the terms you're going on about, so it just looks like a strawman argument to me.

Besides, all that number waffle is entirely academic, since no-one ever faces the choice of (Switching from A to B) OR (Switching from B to C) - they face the choice of (Switching from A to (B OR C))

Re:However, something important to keep in mind

[Rennt]

The 11 MPG improvement switching from the SUV to sedan saves you 3.1 gallons per 100 miles driven, while the 25 MPG improvement switching from sedan to hybrid only saves you 2 gallons.

Classic mistake. You can' t make a comparison without a baseline.

So lets use GPM.
4 / 7.1 = 0.56
2 / 7.1 = 0.28

OK, what about MPG?
25 / 50 = 0.5
14 / 50 = 0.28

Doesn't make a lick of difference if you use X/y or Y/x. Switching to a hybrid will save your nearly twice as much fuel as switching to a sedan would. The numbers don't lie.

Re:However, something important to keep in mind

[hairyfeet]

The problem with SSDs is that the tech isn't really ready for primetime IMHO, unless you just have some money burning a hole and don't care about the data they'll have on them. Atwood at Coding Horror even says SSDs should be judged on a hot/crazy scale since you are dealing with a device that gives serious performance at an INSANE failure rate. From the looks of things they may be full of shit on those MTBF numbers.

All I know is I have a couple of gamer customers for whom the benchmark is God and both went SSD, not cheap shit either, the biggest most expensive they could find. With BOTH drives what happened is one day they flipped the switch and....nothing. That's it, just dead. No warning, no SMART, just tits up DOA bye bye data. Needless to say they weren't happy.

So i think I'll let others blow the crazy money on a drive that may only last a year, since with the big fat caches on the new HDDs combined with plenty of RAM for Windows 7 to superfetch everything seems to me more than fast enough for my customers. I just can't see myself spending that kind of money then having to be drawn up in a knot or do backup after backup just to keep from having to worry about flipping the switch and watching my data go poof. Faster is nice, but not when it is going really fast right off a cliff.

Re:However, something important to keep in mind

[Ant P.]

The metric we should be using is GPM, not MPG.

You mean l/km.

Re:However, something important to keep in mind

[GooberToo]

You notice a major difference between an SSD and a HDD.

I completely agreed. Most report total drive failure in 6-18 months with SSD drives with absolutely no warning of impending catastrophic failure. So ask yourself, are you ready for complete data loss every 6-18 month, while paying an extreme premium for the privileged? Most recommend NOT using these as your primary device, but rather use a smart caching device, which uses these devices as an accelerator while still using the HDD as your primary, dramatically more reliable, long term storage medium. Of course, in either case, a sane, complimentary backup strategy should employed.

Re:However, something important to keep in mind

[hairyfeet]

Well yeah HE says its worth it, you have to remember this is a guy that wears $400 headphones! he really doesn't give a shit if he blows a couple of grand a year on SSDs and says as much in the article I linked to as long as he gets crazy speeds. but you'll notice in the comments while some have had a good experience there are just as many with VERY expensive drives that are now paperweights.

I'm sorry but those kinds of failures not only are unacceptable to the masses, but more importantly the WAY they fail is really unacceptable. In all my years I have yet to have a HDD "just die" without giving SOME kind of warning, be it smart, be it motor whine, be it rising temps...something would clue me in that the customer was in trouble and this gave me time to get the data off. I have had to use Spinrite occasionally when a customer simply didn't heed the warnings and starting throwing errors, but again we are talking about being able to save a good 80%+ of the data.

With the SSDs that my gamer customers picked up it was just...poof. That's it. No warning, no SMART, no errors, they just turned it on one day and found a brick where their SSD used to be. One I managed to get a small amount of data off of, the other? Couldn't even be seen by BIOS.

Frankly I have never seen any PC component just instantly go tits up like that and the fact that data is supposed to be kept on this thing? Uh uh, no way. If you are like Jeff atwood and pay $400 for a pair of headphones and think a couple of grand blown on SSDs is just cool beans as long as you get hot performance? Then play the hot/crazy scale. But for prices of these to truly approach SSD you need the mainstream to accept them and there is no way Joe Average is putting up with having his drive go tits up yearly and take his data with it.

I think my gamer customers had the right idea after getting burned. they just bought a pair of velociraptors and went RAID 0 and then used a big fat 5900RPM drive to do backups. They still get pretty decent performance but don't have to look at pushing the on button as cutting the red wire or seeing their data go float away like a fart on the breeze. If you want to do the same they were selling 74Gb raptors open box for pretty cheap at Newegg. hell of a lot safer than SSDs ATM.

Why no testing with pci-e SDD cards next to the

[Joe_Dragon]

Why no testing with pci-e SDD cards next to the Sata-3 disks?

also should test with a HIGH END SAS / SATA raid card.

Re:Why no testing with pci-e SDD cards next to the

[ub3r n3u7r4l1st]

Right now TRIM command don't pass through RAID chips. So unless these SSD comes with robust garbage collection algorithm I wouldn't put those in RAID.

Re:Why no testing with pci-e SDD cards next to the

[DigiShaman]

I was about to correct you regarding Intel's support for TRIM using RAID. But when I went to double-check, Intel corrected their previous statement.

Intel® Rapid Storage Technology 9.6 supports TRIM in AHCI and RAID modes for drives not part of a RAID volume. A correction was filed to update the information in the Help file, which stated TRIM was supported on RAID volumes.

Solution ID: CS-031491
Date Created: 24-Mar-2010
Last Modified: 05-May-2011

http://www.intel.com/support/chipsets/imsm/sb/CS-031491.htm

Re:Why no testing with pci-e SDD cards next to the

[billcopc]

TRIM is nowhere near as big an issue as people think. GC works fine on both Indilinx and Sandforce units. Realistically, once you start talking about RAID, the modest and highly situational performance gains from TRIM become irrelevant. Even with TRIM support, the block erase still happens in the background, so there is some delay before write performance is restored. If you're in a heavy rewrite scenario, you gain nothing as the drive doesn't idle long enough to handle those deferred erases anyway. And if you're not a heavy rewriter, the GC will take care of it sooner or later. It's all so moot.

My Revodrive X2 hits 650mb/sec with ease. My new Velodrive goes up to about 1000mb/sec, despite having a gazillion apps and games installed and my careless use of my "desktop" folder as temp space. The bottleneck is the shitty Silicon Image RAID chip, not the lack of TRIM. A very interesting product on the radar is the new Angelbird Wings, another PCI-E SSD with some cool features not found in the other brands, like mounting ISOs as virtual CDs at boot time, and a built-in Linux/XFCE distro for management/partitioning. They're claiming 900mb to 1000mb/sec speeds on the 4-channel model.

Still think TRIM is the deal-breaker ?

SATAIII is great, but unstable

[Anonymous Coward]

Before getting excited and rushing off to buy a SATAIII SSD, bear in mind that there are currently some serious stability issues in some of the drives. I recently bought a Corsair Force 3 120GB, only to discover that many, many (as in most) customers are experiencing problems with system hangs and BSODs. I've been lucky enough to only have the drive lock up once a day after a full day of use, but plenty of folks only last about an hour. The old drives were recalled, but even the replacements are having the same problems. This a big support thread at http://forum.corsair.com/forums/showthread.php?t=96333, and plenty more just like it in the same forum.

Corsair seems to think this is an issue with the SandForce controllers, which is reasonable considering other big names (OCZ, for instance) are having the same issues with some of their new drives. Just be sure to do your research first.

Re:SATAIII is great, but unstable

[theshowmecanuck]

I'm not going to rush out and buy any SSD until the price point comes WAY down. Right now my tried and true SATA HDDs work just fine. And I like my wallet performance better when it has more dollars in it because I didn't go right out and buy the newest stuff when the old stuff still does the job. Maybe in a couple more years when the price is better, and when I actually need to replace a drive because it fails or I build a new workstation.

New way of thinking needed

[Archangel Michael]

What is needed is a new way of thinking about memory/storage. More "unix" like thinking where the entire Processro/Cache/RAM/SSD/HDD/Cloud/Tape concept is a singular flat memory space that is addressed as needed. Processor/Cache for instantaneous use, RAM for immediate use, SSD for near RAM fast use, HDD for occasional user, and so on. Where the files (or bits of files) that are needed often are moved closer to the Core processor as needed, automatically.

Re:they still need to be a lot bigger now 500GB an

[fuzzyfuzzyfungus]

While, of course, I'd like all that and a pony, I recognize that ours is not exactly a perfect world. Big SSDs are expensive and big HDDs are (comparatively) slow.

What would be nice, though, and arguably rather more reasonable(it's only a matter of software, and across millions of users the unit cost should be approximately fuck all), would be seeing the tech for transparently dividing workloads across two or more disks with heterogenous characteristics descend from its present position in expensive SANs and comparatively esoteric server FSes.

Sure, the manual "OS+applications on SSD, porn and torrents on HDD" tactic works more or less alright; but having humans wasting their time doing a (lousy) attempt at a machine's job seems like such a pity. Handling the messy details of physical storage location, in order to achieve best apparent performance with lowest burden on the operator, is exactly the sort of abstraction that our computers should be handling for us.

I'll stick with Intel

[vsage3]

I bought one of Intel's 3rd generation 80GB SSDs back in January and have had zero problems with it. No, it's not as fast as OCZ's drives, but it's reliable. Intel's failure rate is 0.6% while OCZ's is 3% (not sure if that's a per-year figure or something). Why an average user would buy primary storage with a 3% failure rate is beyond me.

(failure rate figure comes from http://www.anandtech.com/show/4202/the-intel-ssd-510-review/3 )

Re:I'll stick with Intel

[devleopard]

I burned through 2 OCZ Vertex 3's in like 2 weeks. One ran for a day, then was so screwed the system wouldn't power .. some sort of crazy short. So I took it back, and was lured by the speed, and replaced it with same drive. Was good, but random blue screen. Turns out their firmware is garbage for Win 7 ... supposedly you can update the firmware, but both the Windows approach (had to hook drive up to second machine, as it won't update if it's a boot drive .. WTF) and their Linux boot reported firmware was up to date. (had 2.06 - most recent is 2.09 .. again, WTF)

Took it back, replaced with a "slower" SSD. (One random blue screen means I've lost all those milliseconds over the life of the drive I'm gaining with a faster drive.) At the end of the day, my computer has to work, or I won't be able to.

Re:I'll stick with Intel

[Jeremi]

At the end of the day, my computer has to work, or I won't be able to.

At the end of the day you should go home and relax. Don't burn yourself out!

I'm missing the Intel 510 here

[Skuto]

It's got a SATA-600 interface, the same Marvell controller like the Crucial m4, but it's significantly faster overall. Should be more competitive with the OCZ offerings, and it's 5 times less likely to eat your data.

6 disks, 6 awards

[myspys]

Howcome all 6 disks got an award? Either "recommended" or "editors choice".

Is that the only way to keep everyone happy, and the freebies coming?

Re:6 disks, 6 awards

[after.fallout.34t98e]

Because the editors are still in their heads comparing them to HDDs. It will be a few more years before the wow factor of just how fast these things are winds down and reviewers become able to compare them against each other and not the spinning media.

Re:they still need to be a lot bigger now 500GB an

[Joce640k]

10 GB is easily enough for today's OS with most programs (3 GB or less with some sacrifices),

You haven't tried Windows 7 then? I don't think you can even install it in 10Gb. I'd say 25Gb is a bare minimum for Windows 7 plus a few applications.

Re:they still need to be a lot bigger now 500GB an

[SenseiLeNoir]

I have created a system recently with a 120GB SSD, and a 1 terabyte HDD. I installed Win 7 and some important applications on the SSD, and installed all other applicatiosn as well as the user folders, and swap on the HDD. I get pretty much no difference in performance to a full SSD system, whilst still havign capacity at a price I can afford.

A friend of mine has done a similar thing with Linux and he is getting pretty good results in the performace stakes too.

though performance does drop off ...

[Have Brain Will Rent]

though performance does drop off as the data gets more incompressible."

This is the real problem. Performance figures for SSD's are quoted based on highly compressible data because the SSD compresses the data before storing it thus performing fewer physical writes. Similarly for reading. This gives a highly distorted view of SSD performance. It's like saying "my car gets 200 MPG" and neglecting to say that figure is based on all downhill mileage.

Try copying 10GB of already compressed data and the SSD results are markedly different than what the benchmarks and specifications show. IMHO if this is not clearly pointed out on SSD sales literature then it is tantamount to fraudulent advertising.

In many cases, and for the same price, a couple of software raided hard drives will offer as good as or better performance than an SSD and with an order of magnitude more storage.