The idea of a mSATA-based SSD plugged into your PC's motherboard is cool, but only insofar as the little drive is as fast as more familiar 2.5" models. And it needs to cost the same per gigabyte, too. Paying extra for a slower drive just doesn't make sense, even when it's a small SSD used exclusively for caching. With more and more standard SSDs falling under $1/GB, it's pretty easy to work at least 128 GB of capacity into most desktops.
But in an environment where extra physical space is a luxury (and in some cases not an option at all), mSATA might be the only way to get solid-state storage at all. Ultrabooks are a prime example. With very limited room, you're looking at one 2.5" storage device, an mSATA-based drive, or some combination of the two. Paying a little extra for the functionality of one fast boot drive and a slower disk for user data might be very well be worthwhile in such a compact form factor.

In case you missed Intel SSD 310 80 GB: Little Notebooks Get Big Storage Flexibility, where we first introduced mSATA nearly two years ago, the physical connector we're talking about today looks a lot like mini-PCIe. However, mSATA employs native SATA signaling. Fortunately, some of the boards currently available have the multiplexers necessary to take full-length mini-PCIe cards or mSATA drives in the same slot. From there, you face a performance challenge: some motherboards, such as Intel's DH61AG, enable mSATA at 3 Gb/s transfer rates, handicapping faster SSDs intended for 6 Gb/s connections.
All of that is to say pairing a platform to a compatible mSATA-based SSD is not always an easy exercise. Getting it right, however, can be rewarding. Beyond the interface's purpose as an enabler of caching (which is only of moderate interest to us nowadays), we really do like the idea of 128 GB or more of flash-based capacity and a 500 GB or larger mechanical drive for user data (like music, movies, and pictures).
Seeing that the selection of SSDs designed to drop into mSATA slots is pretty small, we rounded up as many as we could get our hands on from Adata, Crucial, Mushkin, and OCZ.
Although we see Ultrabooks (and other thin and light form factors) as the most likely beneficiaries of mSATA space savings, they don't make great benchmarking platforms for swapping drives in and out. So, we're using an mSATA-to-SATA adapter on our desktop test bed. There is no performance penalty associated with this configuration, since the adapter simply changes the physical interface, not the signaling. As a side benefit, this setup lets us compare mSATA-based drives to larger 2.5" SATA-based SSDs without introducing any other variables.

| Test Hardware | |
|---|---|
| Processor | Intel Core i5-2400 (Sandy Bridge), 32 nm, 3.1 GHz, LGA 1155, 6 MB Shared L3, Turbo Boost Enabled |
| Motherboard | Gigabyte G1.Sniper M3 |
| Memory | Kingston Hyper-X 8 GB (2 x 4 GB) DDR3-1333 @ DDR3-1333, 1.5 V |
| System Drive | OCZ Vertex 3 240 GB SATA 6Gb/s |
| Tested Drives | Adata XPG SX300 64 GB SATA 6Gb/s, Firmware: - |
| Adata XPG SX300 128 GB SATA 6Gb/s, Firmware: - | |
| Adata XPG SX300 256 GB SATA 6Gb/s, Firmware: - | |
| Crucial m4 mSATA 64 GB SATA 6Gb/s, Firmware: - | |
| Crucial m4 mSATA 128 GB SATA 6Gb/s, Firmware: - | |
| Crucial m4 mSATA 256 GB SATA 6Gb/s, Firmware: - | |
| Mushkin Atlas m4 mSATA 60 GB SATA 6Gb/s, Firmware: - | |
| Mushkin Atlas m4 mSATA 120 GB SATA 6Gb/s, Firmware: - | |
| Mushkin Atlas m4 mSATA 240 GB SATA 6Gb/s, Firmware: - | |
| OCZ Nocti 120 GB GB SATA 6Gb/s, Firmware: - | |
| Intel SSD 310 80 GB SATA 3Gb/s, Firmware: - | |
| Intel SSD 320 300 GB SATA 3Gb/s, Firmware: 1.92 | |
| Intel SSD 320 80 GB SATA 3Gb/s, Firmware: 1.92 | |
| Intel SSD 330 180 GB SATA 6Gb/s, Firmware: 300i | |
| Intel SSD 330 120 GB SATA 6Gb/s, Firmware: 300i | |
| Samsung 830 256 GB SATA 6Gb/s, Firmware: CXMO | |
| Samsung 830 64 GB SATA 6Gb/s, Firmware: CXMO | |
| Crucial m4 256 GB SATA 6Gb/s Firmware: 0309 | |
| Crucial m4 64 GB SATA 6Gb/s Firmware: 0009 | |
| OCZ Vertex 3 240 GB SATA 6Gb/s, Firmware: 2.15 | |
| OCZ Vertex 3 120 GB SATA 6Gb/s, Firmware: 2.22 | |
| OCZ Vertex 3 60 GB SATA 6Gb/s, Firmware: 2.15 | |
| OCZ Agility 3 240 GB SATA 6Gb/s, Firmware: 2.22 | |
| OCZ Agility 3 120 GB SATA 6Gb/s, Firmware: 2.22 | |
| OCZ Agility 3 60 GB SATA 6Gb/s, Firmware: 2.22 | |
| OCZ Vertex 4 256 GB SATA 6Gb/s, Firmware: 1.5 | |
| OCZ Agility 4 256 GB SATA 6Gb/s, Firmware: 1.5 | |
| OCZ Agility 4 128 GB SATA 6Gb/s, Firmware: 1.5 | |
| OCZ Vertex 4 64 GB SATA 6Gb/s, Firmware: 1.5 | |
| Graphics | Palit GeForce GTX 460 1 GB |
| Power Supply | Seasonic 760 W, 80 PLUS Gold |
| System Software and Drivers | |
| Operating System | Windows 7 x64 Ultimate |
| DirectX | DirectX 11 |
| Driver | Graphics: Nvidia 270.61 RST: 10.6.0.1002 Virtu: 1.1.101 |
| Benchmarks | |
|---|---|
| Tom's Hardware Storage Bench v1.0 | Trace-Based |
| Iometer 1.1.0 | # Workers = 1, 4 KB Random: LBA=8 GB, varying QDs, 128 KB Sequential |
| PCMark 7 | Storage Suite |

Adata's XPG SX300 family is the mSATA-based relative of the company's 2.5" SX900 line-up, which we covered in Time To Upgrade: 10 SSDs Between 240 And 256 GB, Rounded Up. They both employ SandForce's second-generation controller, suggesting that we may just see the compact form factor serving up performance similar to what we've been enjoying on the desktop for more than two years now.
In reality, the XPG SX300s are a little different than the most conventional SandForce-based drives. Like the SX900s, Adata takes advantage of a vendor update from the controller company that allows it to disable over-provisioning entirely. In our review of the SX900, we saw that over-provisioning helps the drive recover performance when all of its cells are written to. Getting rid of the feature can affect speed then, under the right conditions.

You'll notice that the XPG SX300s (all three of them, at 64, 128, and 256 GB) host four BGA memory packages, each of which connects to the controller via two channels. Thus, all three mSATA-based SSDs communicate over eight available channels. Adata tells us that it's using 25 nm synchronous flash from IMFT, which should enable impressive performance.
Why is there so much difference between the three versions in our 4 KB random read test, then? At the low queue depths you'd expect to encounter on an Ultrabook or mainstream desktop, the drives are actually fairly close together. It's only at queue depths of eight and more that the 256 GB model really takes off. This is a result of the interleaving that takes place as more NAND devices are involved in a given operation. At high queue depths, the larger SSD is able to better-utilize the interface between SandForce's controller and the attached flash.

Writing compressible information to the XPG SX300s (in the chart below) doesn't leave any room for interleaving to help performance. SandForce's DuraWrite technology helps all three SSDs achieve similar results.
However, we know that SandForce's architecture doesn't handle compressed data as elegantly. So, the 128 and 256 GB drives serve up notably better performance than the 64 GB model, which is composed of few NAND die. Moreover, the trio's benchmark numbers are significantly lower when each drive is forced to contend with incompressible information.

Sequential read performance is notably better. All three drives post similarly-impressive throughput, sailing over 500 MB/s at a queue depth of two.

So long as you're dealing with compressible data, 128 KB sequential write performance is also really impressive.
The impact of interleaving is profound once we force these mSATA-based SSDs to cope with purely incompressible information. The 128 and 256 GB models fall to 200 MB/s, while the 64G GB drive drops to 100 MB/s. Moreover, higher queue depths do nothing to improve transfer rates, since the architecture is already saturated at a queue depth of one. Stacking more operations only yields better performance when there's still headroom available.


Crucial's nomenclature suggests that we're dealing with a pared-down version of the company's 2.5" m4. The mSATA-based family is available in 32, 64, 128, and 256 GB capacities, though we're only testing the three largest models today, and probably wouldn't recommend bothering with a 32 GB drive anyway. That's barely large enough for a 64-bit Windows 7 or 8 installation, which eats up more than 20 GB on its own.

Crucial's m4s with mSATA interfaces employ the same eight-channel Marvell 88SS9174 controller found on the company's older and more familiar 2.5" models. The company also uses the same 25 nm NAND from IMFT. As you can see, though, there is only room for four BGA memory packages on the diminutive PCB. So, some of the same interleaving issues that affected Adata's XPG SX300 drives are bound to surface when we compare lower-capacity drives to the larger models with more flash.

Random read performance is consistent from a queue depth of one all the way through 32, with the 256 GB m4 trailing just slightly. This is in contrast to what we saw from Adata's drives, which were clearly differentiated by the time we hit higher queue depths.

Without the benefit of DuraWrite, a SandForce-only technology, Crucial's m4s behave a lot like Adata's XPG SX300s faced with incompressible data. They all turn in similar numbers at a queue depth of one in our 4 KB random write test. At a queue depth of two, the 128 and 256 GB models stand apart from the 64 GB drive. And by the time we push a queue depth of four, the 256 GB sits atop Crucial's stack.

It takes a queue depth higher than four for the 64 and 128 GB m4s to catch up to Crucial's 256 GB model, though, at several other points in the chart, all drives perform similarly in our 128 KB sequential read test.

DuraWrite helped Adata's XPG SX300s maintain comparable transfer rates in our 128 KB sequential write benchmark, so long as data was compressible. As soon as the SandForce-powered SSDs were faced with incompressible data, however, we saw clear differences between each capacity level.
It doesn't matter what type of data you throw at Marvell's controller; the m4s don't exhibit that dualistic behavior typical of SandForce-based competition. But right out of the gate we see the 64 GB m4 achieving less than half as much throughput as the 256 GB model, with 128 GB between them.

Unlike Adata's XPG SX300 family, Mushkin's Enhanced Atlas line-up is available in three more familiar (at least for SandForce-based SSDs) capacities: 60, 120, and 240 GB. The Enhanced Atlas centers on the same second-gen controller, but Mushkin maintains over-provisioning on its models to help improve the speed with which these drives recover their performance, even when they're full.

Based on that distinction alone, we'd suspect the Atlas drives to be more performance-oriented than Adata's XPG SX300s. However, we also see that Mushkin arms its SSDs with Toggle-mode NAND from Toshiba. Based on that fact, these things should be quicker (at least in theory) than what we've already seen from Adata using the same controller.

Unfortunately, our first test, 4 KB random reads, don't support our supposition. All three models perform slightly worst than what we saw from the XPG SX300 family. And while we might have expected the impact of interleaving to lend the 120 GB drive better numbers than the 60 GB configuration, it's only at a queue depth of 32 that it differentiates itself. Mushkin's 256 GB version, in contrast, is faster at every queue depth.

It should come as no surprise that the Enhanced Atlas line-up yields results similar to Adata's XPG SX300s in our 4 KB random write test. If you hit these SSDs with compressible data, they're super-fast at high queue depths (though not notably more than the Adata models with synchronous NAND).
Unfortunately for Mushkin, performance in the face of incompressible data is actually worse from its Atlas drives than the competition from Adata. What makes this unfortunate? Well, as the fastest NAND interface available Toggle-mode-capable flash is supposed to be the most expensive. Technically, you'd pay more for a Mushkin Atlas with less usable capacity than a comparable Adata XPG X300. However, a quick spot check on Newegg reveals that, although Muskin's 60 GB model is more expensive, as we'd expect, its 120 and 240 GB SSDs actually cost less than the competition (by $20, in the case of the 240 GB drive).

Toggle-mode NAND does facilitate slightly better numbers in our 128 KB sequential read test, compared to the synchronous-equipped XPG SX300s. At all three capacity points, the benchmark results are similar.

That's not the case for sequential writes, though, where the Enhanced Atlas drives are a little slower than Adata's models when faced with compressible data.
Turning to incompressible information, the 120 and 240 GB models are slower than what we saw two pages ago, while Mushkin's 60 GB SSD is actually a bit faster than the 64 GB XPG SX300. Notice that all three Enhanced Atlas drives fall much closer to each other than the Adata versions did.

OCZ sells a great many 2.5" SSDs, but it only has one mSATA offering called the Nocti. You'll want to keep in mind that it's an older design, which is why it's specifications are mismatched against the Adata, Crucial, and Mushkin contenders in our round-up. Nevertheless, you can still buy Nocti drives online. In some cases they're even more expensive than the competition. So, it's important that we include the Nocti in our benchmark results.
Notice that OCZ is using a second-generation SandForce controller on its sole mSATA-based SSD. On paper, that'd give it the chops needed to hold its own in testing.

However, we can also see that the company is leaning on TSOP packaging. Unlike the BGA modules used by Adata, Crucial, and Mushkin, these TSOP packages communicate over a single channel each. An mSATA drive with four packages is only utilizing half of what the SandForce controller has available. And that's why OCZ only rates the drive for sequential reads of up to 280 MB/s and writes as high as 260 MB/s.
Now you see just how dire the Nocti's disadvantage is. That it's still selling for as much as some of the competing drives out there really doesn't make sense.

In lightweight storage workloads, where the Nocti is faced with a queue depth of one, it's actually just as fast as the higher-end SSDs we're testing. It's even competitive at a queue depth of two. But when you start stacking I/O operations, it quickly falls behind. Fortunately for OCZ, most mobile and mainstream desktop applications are not storage intensive. Even still, at the same price (or less), we'd rather spring for the SSD able to utilize all eight of the SandForce controller's channels.

Writes (particularly writing incompressible data) are dismal compared to the Adata and Mushkin drives. Whereas the 120 GB Nocti tops out around 80 MB/s, those two competitors exceed 150 MB/s.


Sequential reads and writes are also pretty bad in comparison to the XPG SX300s and Enhanced Atlases.
At least, that's in theory. Iometer workloads are fairly synthetic, mostly useful for validating manufacturer specifications. Moving on to more real-world testing should help us determine whether OCZ's spec sheet-based disadvantage is something you'd need to worry about in day-to-day use.

Crucial's m4s deserve special recognition. The 128 and 256 GB versions are almost exactly as fast as each other; that's notable because competing 120 and 128 GB models are a bit slower. And the 64 GB m4 holds its own against larger competing SSDs.
All discussion of other drives is rendered almost needless by the fact that Crucial's 256 GB m4 currently sells for $60 less than its closest competitor. It is worth pointing out, however, that OCZ's Nocti drags along at the bottom of our chart.

That 256 GB m4 gets even more attractive when you take a look at its power consumption under load. It ducks in under 1 W, lower than any other mSATA-based SSD in today's round-up and followed by the other two m4s.
At the other end of the spectrum, higher-capacity drives with SandForce controllers use notably more power under load. The Adata XPG SX300 and Mushkin Enhanced Atlas families don't fare particularly well.
They're all pretty close together at idle, with the exception of OCZ's Nocti, which uses very little power indeed.

Iometer is a great tool for helping us isolate the behavior of storage devices in very specific cases. But simply running 4 KB blocks of random data or 128 KB chunks of information sequentially through a drive doesn't give us a true sense of performance in a real-world environment. And that's because most real-world workloads aren't nearly as single-dimensional.
So, we also have a couple of real-world file transfer tests, both of which are indicative of write performance as we move information off of a 240 GB Vertex 3 and onto each of these mSATA-based SSDs.

The first test involves one large media file, which is incompressible due to its already-encoded container. It might surprise you to see so much differentiation between the various drives. However, at the top end, Crucial's m4 isn't affected by the type of data we're moving. And, at the bottom, Intel's SSD 310 only populates five of its controller's 10 available channels, helping explain why it doesn't fare as well.
In fact, we can do a bit of math to figure out just how well each of these SSDs matches up to its rated specifications. Intel claims the SSD 310 80 GB can sustain 70 MB/s sequential writes. We show it doing 79 MB/s. Crucial says the 256 GB m4 should be able to write at up to 260 MB/s. We have it pegged at 244 MB/s. Overall, these numbers look pretty darned close.
So yes, while the responsiveness of an SSD is universally better than what you get from a hard drive, there can be some pretty substantial differences in performance when it comes to moving around lots of information.

Here, we're writing a large folder of files that blend random and sequential accesses. Much of this data is already compressed into large files by Blizzard, so we're not sure how much more SandForce's DuraWrite technology can do to improve performance.
Now we see Intel's SSD 310 doing 107 MB/s, while Crucial's m4 smokes along at 340 MB/s. Both of those figures are in excess of what each vendor rates its product for. However, we ran each drive twice to be sure they were correct and saw consistent results within four seconds from one transfer to the next.
Notable yet again is how much worse the 60/64 GB models do compared to the 120/128 and 240/256 GB drives. That information is particularly important as we come up with some general recommendations on the next page.
The first mSATA-based SSDs that emerged frankly didn't do much to impress us. It was cool to get our hands on a device the size of a mini-PCI Express card with, at the time, 80 GB of capacity (Intel SSD 310 80 GB: Little Notebooks Get Big Storage Flexibility). However, smaller dimensions translated directly to compromises in performance, since Intel was only able to populate five of its controller's 10 channels using TSOP memory.

Today, the story is a lot different. Higher memory density and BGA packaging make it possible for vendors to fully utilize some of today's fastest storage controllers. And the fact that we're seeing 256 GB drives means you don't have to lean on mSATA-based SSDs for caching; they're able to stand in as a primary target for your operating system, performance-sensitive applications, and games.
And then there's pricing. At launch (and even today), Intel wanted $180 bucks for its SSD 310 80 GB, a drive we see bringing up the rear of our benchmarks. Now, you can find a 256 GB Crucial m4 for $200. That's well under $1 per gigabyte, and exactly as much as the company's 2.5" version of the same drive. Gone are the premiums on mSATA-based drives (at least, the ones we'd recommend to you).
| Cost Breakdown | Market Price | Price Per GB |
|---|---|---|
| Adata XPG X300 64 GB | $90 | $1.41 |
| Adata XPG X300 128 GB | $130 | $1.02 |
| Adata XPG X300 256 GB | $270 | $1.05 |
| Crucial m4 mSATA 64 GB | $70 | $1.09 |
| Crucial m4 mSATA 128 GB | $115 | $0.90 |
| Crucial m4 mSATA 256 GB | $200 | $0.78 |
| Intel SSD 310 80 GB | $180 | $2.25 |
| Mushkin Enhanced Atlas 60 GB | $75 | $1.25 |
| Mushkin Enhanced Atlas 120 GB | $120 | $1.00 |
| Mushkin Enhanced Atlas 240 GB | $210 | $0.88 |
| OCZ Nocti 120 GB | $130 | $1.08 |
In fact, breaking down the cost of each drive tested today, eight of the 11 models sit around $1/GB or less. The notable exceptions are Adata's XPG 64 GB, Intel's SSD 310 80 GB, and Mushkin's Atlas 60 GB. Incidentally, based on what we saw in the benchmark results, we'd recommend against all of those models. And although its price per gigabyte is a little more attractive, we'd add Crucial's m4 64 GB to that list for its low capacity and modest performance. We'd also add OCZ's Nocti 120 GB for its performance and price proximity to much more attractive options.
The 120 and 128 GB SandForce-based drives are pretty appealing. Adata's XPG SX300 and Mushkin's Enhanced Atlas 120 GB are both strong contenders. So too are the 240/256 GB versions of those same SSDs. But we're most eager to single out Crucial's 256 GB m4. Yes, the 128 GB version is a little more accessible at $115. However, the 256 GB model enjoys the lowest cost per gigabyte of any SSD in our comparison, its power consumption is well within where we'd want to see it, and performance remains impressive, despite the age of its architecture.

I never had an opportunity to give one of our Best of Tom's Hardware awards, and we're revising our awards somewhat to better reflect their true intention. So, it's a privilege to hand out our first Tom's Hardware Elite award to Crucial's 256 GB m4 in the mSATA form factor. This recognition is deserved because the drive demonstrates consistent performance at a price we're comfortable paying. It's unclear if we'll ever see a dramatically faster SSD for the mSATA interface. But, seeing as this one already comes close to saturating a 6 Gb/s connection in sequential reads, and is well-established as a reliable solution, there isn't another option we'd recommend as an alternative. The 256 GB m4 takes our highest honor, and is the drive to buy if you need mSATA-based storage.
