by Ilya Gavrichenkov
09/10/2011 | 04:59 PM
SandForce’s is a story of success on the semiconductor market. A small team of experienced engineers in a very short time developed and put into production a solid state drive platform which has become one of the most popular among SSD makers as well as computer enthusiasts. There seem to be two factors contributing to that success. First, SandForce came up with an innovative technology that helped squeeze a very high performance out of MLC NAND flash memory. Second, this technology was presented in a manufacturer-friendly way. SandForce do not offer a bare controller. Instead, they offer a PCB design and firmware which can be tailored to the specific SSD maker’s needs.
As a result, there are now dozens of brands producing SSDs based on SandForce controllers. It is in fact very easy to enter the SSD market with SandForce's help. You don't have to have your own semiconductor fabs or advanced R&D resources. You don't even need programmers. All you have to do as a SandForce partner is to buy flash memory chips, solder them to PCBs, put the PCBs into cases, package the resulting products and sell them through your distribution channels. The profit rate of such a business is still very high, the SSD market developing exponentially. No wonder that besides world-famous brands like Intel, Samsung and Micron, we can see a lot of other, often obscure, brands producing their SSDs. In fact, other than the first-tier brands, SSD makers are all adherents of the SandForce architecture.
We might expect the market to be flooded with clone products, but that’s not really so. The SandForce architecture is flexible. There are several controllers with somewhat different basic functionality. Each SandForce controller is compatible with different varieties of NAND flash memory irrespective of the latter’s interface, speed and price. For example, the SF-2000 series controllers can work with new 25nm NAND flash as well as with 3x-nanometer chips, with both MLC and SLC memory, and with different interfaces (synchronous or asynchronous, ONFI or Toggle Mode). The consequence is that even the latest generation of SSDs with SATA 3.0 (SATA 6 Gbit/s) interface that are based on the same SF-2000 controllers can differ greatly in their consumer properties, including price.
For example, based on the same modification of the SandForce SF-2281 controller, the 120GB OCZ Vertex 3 Max IOPS costs $270, while Corsair Force 3 with the same capacity – only $180, and this difference in price surely reflects a difference in performance. It means that your shopping choice should be based not only on what controller the particular SSD has but also on other factors since SSD makers can change the speed of their products by choosing different flash memory or optimizing firmware. In other words, if you like the OCZ Vertex 3 we tested earlier, it doesn’t mean that every other SATA 6 Gbit/s SSD with a second-generation SandForce controller is going to be just as fast. That’s why we are going to carry out a comparative test of new SandForce-based products with the latest SATA interface which are currently available aplenty in shops.
In this review we will take a look at seven products from Corsair, Kingston, OCZ and Patriot.
Each SandForce controller implements an original concept of compressing data “on the fly” prior to writing it into flash memory. Although this makes the controller itself somewhat more complex, the benefits are indisputable. The compression of data helps reduce the number of flash memory accesses to increase performance. Another consequence of compression is that the service life of an SSD is increased since there are fewer rewrite cycles. SandForce controllers also feature RAISE technology which uses some of the available flash memory for storing checksums which help recover data in case of read errors.
In other words, the SandForce architecture is well-suited for inexpensive NAND flash of inferior quality and performance. Even if the SSD uses asynchronous chips with ONFI 1.0 interface and a bandwidth of 50 MB/s, the eight-channel SF-2281 controller helps achieve acceptable speed: up to 400 MB/s with non-compressible data. And if data can be compressed, which is typical of desktop applications, the data-transfer rate may get much higher and even exceed the bandwidth of the SATA 6 Gbit/s interface. Thus, the SSD manufacturer can look for the optimal balance between performance and price, delivering very different products.
Taking the products we’re going to test today as an example, let’s see how many SSD modifications can be built on the basis of one and the same SandForce SF-2281 controller.
So we can see that these SSDs use very different NAND flash types, some of which are 3 or 4 times as fast as others. However, the manufacturers are not eager to specify real-life characteristics for their products. Thanks to the SF-2281’s ability to compress data written into memory, they prefer to talk about peak performance which depends on the controller only.
We’ve got a paradoxical situation as the consequence: SSDs with almost identical official specs may differ by 50% in price due to differences in their internal design. Obviously, they can’t deliver identical performance. The benefits of faster flash memory should show up somehow.
Indeed, the choice of flash memory affects the internal speed of an SSD. Synchronous flash memory not uses a stobe signal to transfer data, but is at least twice as fast as asynchronous flash memory thanks to DDR mode support. Therefore synchronous-flash SSDs have fewer bottlenecks and can deliver high performance irrespective of whether data can be compressed or not (archives, JPEG images, H.264 videos and other such files are almost completely incompressible). The downside is that such SSDs are more expensive.
The cheaper asynchronous variety of flash memory limits the data transfer speed between the controller and flash memory chips even if they are connected through as many as eight channels. When there is a high incoming data stream that can't be reduced by compression, the overall performance of the SSD degrades.
In other words, expensive synchronous NAND flash with a bandwidth of 166 or 200 MB/s has practical benefits. This is the reason why SSDs with ONFI 2.0 or Toggle Mode NAND flash are positioned as premium products. Asynchronous-flash SSDs are more affordable.
Now that we’ve found out the reasons for the different positioning, let’s take a look at each of the products we are about to test.
The Corsair Force 3 is one of the cheapest SSDs in this review. It offers 1 gigabyte of SSD storage for about $1.5. Notwithstanding its low price, it represents a full-featured second-generation SandForce platform with an SF-2281 controller and SATA 6 Gbit/s interface.
It is the flash memory chips that are the reason for the low pricing of this SSD. Corsair picked low-cost asynchronous flash memory with a bandwidth of 50 MT/s (megatransfers per second) for the Force 3 series. Our sample contained 16 Micron 29F64G08CBAAA chips although some batches of this product come with Intel’s NAND flash with the same specs. The chips have a capacity of 64 gigabits each and are manufactured on 25nm tech process. The total capacity is 128 gigabytes but the SandForce controller uses some of it for over-provisioning and for storing checksums. Each of the SF-2281 controller's channels is linked to two chips to enable interleaving.
A special feature of Corsair's SandForce-based products is their modified firmware. Corsair optimizes the reference firmware provided by SandForce and uses different version numbers. Corsair’s latest firmware (version 1.3) is based on the SandForce 3.20 firmware which claims to be more stable than the earlier versions.
Besides the inexpensive product with asynchronous NAND flash discussed above, Corsair offers flagship SSDs with a much higher price per gigabyte, up to $1.9. However, the Force GT series is somewhat cheaper than the premium SSDs from other brands but it is impossible to determine the reasons for this positioning from the product specs. Corsair specifies the highest sequential and random data access speed and reports standard information about the internals: SF-2281 controller and SATA 6 Gbit/s. The only distinguishing feature of the Corsair Force GT is its bright red metallic case, but it can hardly affect its performance.
To find out the real difference between the Force GT and Force 3 series we have to look at what’s inside. The PCB of the faster SSD is almost the same but the memory chips have different marking and manufacturer. The Force GT uses Intel 29F64G08ACME2 chips which are manufactured using 25 nm process and contain one NAND device in each package. There are 16 chips, each with a capacity of 64 gigabits, for a total of 128 gigabytes. The point is that the Force GT’s memory is synchronous and has a data-transfer rate of 166 MT/s.
Like in the Force 3, the controller uses eight memory channels with interleaving, so the practical difference between Corsair SSDs of different series can be dramatic. We just have to measure their speed while writing data the SF-2281 controller cannot compress. In this case, the slower flash memory of the Force 3 is going to show up as a bottleneck.
Corsair uses unified firmware for its SSDs, so the Force GT has the same firmware version 1.3 (based on the reference SandForce 3.20 firmware) as the Force 3.
Kingston took a long time to release its SATA 6 Gbit/s SSDs and finally came up with the HyperX series based on the SF-2281 controller. This is one of the most expensive SandForce-based product series available as Kingston asks about $2.0 for each gigabyte of its SSD storage.
What’s the reason for such ambitions? We don’t think it has anything to do with the pretty case with embossed letters and different materials. It can hardly be due to the accessories, even though Kingston offers more of them than any other maker: brackets for 2.5" to 3.5" conversion, a screwdriver, a container for using the SSD as an external USB disk, and Acronis software for cloning HDD data.
Of course, the main thing is concealed inside the SSD case but, unfortunately, we couldn’t take the drive apart due to its nonstandard screws. As far as we know, Kingston installs 25nm synchronous NAND flash from Intel with ONFI 2.2 interface. Each Intel 29F16B08CCME2 chip contains two semiconductor dies and has a capacity of 128 gigabits. Thus, the 240GB SSD has 16 such chips or 32 flash devices. The HyperX owes its high performance not only to the fast flash memory type but also to the fact that the eight-channel SF-2281 controller supports 4-way interleaving. That’s the reason why the Kingston HyperX can be viewed as one of the most technically advanced SandForce-based products.
The HyperX uses the reference SandForce 3.20 firmware.
OCZ is one of the closest SandForce partners and was the first to release products with the SF-2281 controller after its introduction. Right now, the company offers as many as four such product series: Vertex 3 Max IOPS, Vertex 3, Agility 3, and Solid 3. So, the Agility 3 SSD is not the most affordable offering from OCZ although the price per gigabyte of that model is $1.5. The Agility 3 and Solid 3 do not differ much in price, though, so you can’t save a lot of money by preferring the latter.
There is inexpensive asynchronous flash memory inside the Agility 3. Our sample contains 16 Micron 29F128G08CFAAA chips, each with two 25nm semiconductor dies and a capacity of 128 gigabits. So, notwithstanding the low bandwidth of asynchronous memory, the eight-channel SF-2281 controller can enable 4-way interleaving to hide the low performance of the NAND devices.
The Agility 3 offers a total of 256 gigabytes of storage but, like in any other SandForce-based SSD, some of it is allotted for RAISE technology and over-provisioning. 240 GB drives (in reality the actual storage capacity of these drives is 224 "real" gigabytes) have 14.5% of such "hidden" memory, which is exactly the same as in SSDs of smaller size. The ratio is the same with the smaller-capacity products.
OCZ optimizes SandForce’s reference firmware and develops its own software for SSDs. The latest version of OCZ’s firmware is 2.11 but it’s based on SandForce 3.20. OCZ also offers the Toolbox utility which helps you update the firmware, view the SMART data, and perform Secure Erase to restore the SSD's performance to its original level.
The ordinary Vertex 3 was already tested in our labs and we were pleased with its performance. This time around OCZ offers its fastest SATA 3.0 product for us to test. It is called Vertex 3 Max IOPS. The names are similar, suggesting that the Max IOPS has optimized firmware or something like that, yet the considerable difference in price indicates something more fundamental.
Indeed, the internals of the Vertex 3 Max IOPS are dramatically different from anything we’ve seen so far. The key feature of that model is that OCZ put 32nm flash memory from Toshiba with Toggle Mode DDR interface into it. The memory chips of our sample were marked as Toshiba TH58TAG7D2FBAS9. These are 128-gigabit chips containing four 32-gigabit semiconductor dies each. Thus, although the bandwidth of Toggle Mode DDR MLC flash memory is only 133 MT/s, OCZ hopes to increase the performance by means of 8-way interleaving since there are as many as 64 NAND devices connected to the eight channels of the SF-2281 controller.
Another advantage of using old 32nm memory is its higher reliability. Standard 25nm MLC NAND flash can be rewritten about 3000 times before it becomes unable to store data. The older 32nm memory supports up to 5000 rewrite cycles. It means you can write about 860 terabytes of data to a 240GB Vertex 3 Max IOPS until the end of its service life (this theoretical number is based on JEDEC’s formula).
Every OCZ product based on a second-generation SandForce controller has the same unified firmware. So, the latest firmware for the Vertex 3 Max IOPS is version 3.11 and is based on the SandForce 3.20 code which claims to be more stable than the earlier versions.
Inexpensive SSDs based on second-generation SandForce controllers are highly popular among computer users. They are but slightly inferior to their premium-class counterparts in official specs whereas their price is very attractive. Therefore many companies offer both expensive and affordable SF-2281 based products with SATA 6 Gbit/s. The Pyro is Patriot’s affordable solution whose cost per gigabyte is about $1.6.
As you can guess from its price, the Pyro contains inexpensive asynchronous MLC NAND flash with a bandwidth of only 50 MT/s. Our sample has 16 128-gigabit Intel 29F16B08CAME1 chips each of which has two 25nm NAND devices. Intel and Micron have a joint venture for producing flash memory but the two companies mark their chips differently. In fact, the Pyro has the same chips as the OCZ Agility 3, so the two are identical in terms of hardware (except for certain differences in their PCB wiring).
It doesn’t mean we expect them to perform in the same way, though. OCZ optimizes firmware for its SSDs whereas the Pyro uses the reference version 3.20 firmware.
After the most original Vertex 3 Max IOPS we thought OCZ was the only manufacturer to dare to use 34nm flash memory. However, Patriot has come up with a similar product called Wildfire. It has a SandForce SF-2281 controller and uses MLC NAND flash with Toggle Mode DDR interface. As a result, the Wildfire is just as expensive as the OCZ Vertex 3 Max IOPS and offers 1 gigabyte of storage for about $2.25.
Well, we can’t really view the flagship SSDs from OCZ and Patriot as identical products. First of all, the Patriot is a 120-gigabyte drive whereas OCZ offers a 240-gigabyte one. There are also significant differences inside. The Patriot has different memory chips: Toshiba TH58TAG6D2FBA49 (64 Gb, two semiconductor dies in each). The 16 chips allow the eight-channel SF-2281 controller to enable 4-way interleaving.
Thus, the Wildfire is potentially slower than the Vertex 3 Max IOPS but we don’t think the gap is going to be large since both use flash memory with fast synchronous interface. By the way, the top-end SSD from Patriot boasts increased reliability just like the OCZ product.
Like the Pyro, the Wildfire uses the reference version 3.20 firmware. The manufacturer must have decided that there is nothing to improve in it.
For our SSD tests we put together a special test platform based on Intel H67 chipset, which, as you know, supports a pair of SATA 6 Gbps ports. We are going to use these particular ports for our tests of all second-generation SandForce based SSD, which can take advantage of this high-speed data transfer mode.
Overall we used the following hardware and software components.
We test the random and sequential read and write speed with CrystalDiskMark 3.0.1b. This benchmark is handy as it can measure the speed of an SSD with both incompressible and compressible data. So, there are two numbers in the diagrams that reflect the maximum and minimum speed. The real-life performance of an SSD is going to be in between those two numbers depending on how effectively the SF-2281 controller can compress the data.
We’ve got some interesting results here. We can see that the SSD makers do not mislead us much when they specify similar speeds for very different products. They just talk about the data transfer speed that the SF-2281 controller can compress well. We can’t see any leaders here because the expensive SSDs with synchronous flash are hardly any faster than their cheaper opponents.
Everything becomes clear as soon as we use incompressible data. The more expensive SSDs do not slow down much whereas the SSDs with asynchronous NAND flash deliver only half the read speed they had with compressible data. The standings in the writing test are even more illustrative. First place goes to the Kingston HyperX which is followed by the OCZ Vertex 3 Max IOPS. There are some surprises, though. Some SSDs with asynchronous flash are unexpectedly fast at writing. For example, the OCZ Agility 3 is ahead not only of the Corsair Force GT but also of the Patriot Wildfire.
Thus, the real performance of SSDs based on the SandForce may vary depending on whether data it’s dealing with can be compressed or not. That’s how the performance of a synchronous-flash SSD (OCZ Vertex Max IOPS) depends on the degree of data compression:
And here is the same correlation for an SSD with slower flash memory (Patriot Pyro):
In fact, fast NAND memory sub-system allows creating SandForce-based SSDs, which have constant read speed on all types of data. As for the write speed on incompressible data, it drops anyway, but the drives with synchronous flash inside suffer a much more obvious performance hit.
Now let’s check the SSDs out in random read/write tests.
The four SSDs with synchronous flash (Corsair Force GT, Kingston HyperX SSD, OCZ Vertex 3 Max IOPS and Patriot Wildfire) are somewhat faster than others at random reading. It’s not so simple with random writing: the Patriot Pyro, OCZ Agility 3 and Patriot Wildfire are the best with compressible data whereas the Corsair SSDs are ahead with incompressible ones. This must be due to firmware optimizations rather than hardware features of the SSDs.
The random-address requests to read 4KB data blocks with a request queue depth of 32 are a higher load on the SSDs, so hardware features are as important as firmware optimizations here. As you can see, the OCZ Vertex Max IOPS is the leader in read speed, the Kingston HyperX and Patriot Wildfire being unable to match its performance.
When writing random-address 4KB data blocks with a request queue depth of 32, the SSDs are overall closer to each other than at reading. The Kingston HyperX is in the lead, but we want to note the good performance of some of the SSDs with asynchronous flash. The OCZ Agility 3 and the Patriot Pyro are quite competitive against their more expensive opponents, for example. The two Corsair products are rather disappointing in the tests of random reading and writing, probably due to their modified firmware.
As we’ve just seen, our SSDs can differ greatly in performance in quite a lot of tests. When it comes to compressible data, you can see the SSDs differ in random reading with a long command queue depth. When processing incompressible data, the SSDs differ nearly everywhere, except for random writing. So, whatever marketing claims the manufacturers can make, SSDs with second-generation SandForce controllers are going to differ more in real-life applications than you can expect after looking at their specs. As for the situations when their performance levels are very close, they do exist, although they are fairly rare and you will very unlikely ever experience them.
IOMeter is an industry standard benchmark for disk drives including SSDs. We use its latest version: 1.1.0 RC1. This is a very powerful benchmark with broad functionality but we will focus on the four characteristics which are the most important for SSDs. These are the random and sequential read and write speed.
The first test is about the speed of reading and writing random-address 4KB data blocks with a request queue depth of 1, but we use four independent data streams, one for each CPU core. This disk access scenario is closer to real-life computers running multitask OSes. We use pseudo-random data for this test.
So, it is the OCZ Vertex Max IOPS and the Kingston HyperX that are superior at reading. They are joined at writing by two inexpensive products: OCZ Agility 3 and Patriot Pyro.
The second IOMeter test is the same as the first but the queue depth is increased to 32 requests. This means a much higher load on the disk subsystem.
The first diagram (random reading) shows the most important results because SSDs are going to be used like that in most real-life applications. The OCZ Vertex Max IOPS enjoys a large advantage over its opponents here. When it comes to writing, the Kingston HyperX is in the lead, followed by the OCZ Agility 3 and Patriot Pyro.
The last test is about accessing the disk for sequentially located 128KB data blocks. Like in the previous test, we use pseudo-random data.
IOMeter reports that every SSD based on a second-generation SandForce controller delivers the same performance at sequential reading. As for sequential writing, the Kingston HyperX is ahead while the inexpensive OCZ Agility 3 and Patriot Pyro are again quite fast.
PCMark 7 incorporates an individual disk subsystem benchmark which is based on real-life applications. PCMark 7 reproduces typical disk usage patterns and measures how fast they are performed. Moreover, the disk access commands are not reproduced one by one, but with pauses necessary to process the data, just like in real life.
The benchmark reports an overall disk subsystem performance rating as well as the data-transfer rate in particular scenarios. Take note that the speed in these scenarios is rather low due to the pauses between input and output operations. In other words, PCMark 7 shows you the speed of the disk subsystem from the application’s point of view. Rather than the pure performance of the SSDs, this benchmark will show us how good they are in practical tasks.
The PCMark 7 ratings separate the expensive SSDs with synchronous memory from the cheaper ones with asynchronous MLC NAND flash. The OCZ Vertex 3 Max IOPS is the best of the high-performance products while the OCZ Agility 3 and Patriot Pyro, among the affordable ones.
On the other hand, we can see that the fastest and slowest SSDs differ by less than 10%, which reflects their real-life performance. You may want to take a look at the individual test scenarios to get a full picture.
We test file copy speed using AS SSD version 1.6.4237.30508. The benchmark software copies files within a single partition the size of the SSD’s full storage capacity.
The Kingston HyperX SSD and OCZ Vertex 3 Max IOPS deliver the highest performance when copying files. They are the overall leaders of this test session but they are also expensive. Among the cheaper products, the OCZ Agility 3 and Patriot Pyro should be singled out.
We have tested seven SSDs based on the same SandForce SF-2281 controller and we can conclude that the controller is not the decisive factor that determines the performance of a modern SSD. Firmware optimizations and the choice of NAND flash memory are just as crucial as the controller.
Although they are all based on the SF-2281, no two of them are alike. Each product has peculiarities of its own. We’d say that the main distinguishing factor is the type of flash memory. SSDs with synchronous MLC NAND flash are generally faster but also more expensive whereas SSDs with asynchronous flash are cheaper and slower. However, there are winners and outsiders in each of these two categories, too.
OCZ, being a close partner of SandForce, offers a very wide range of SF-2281 based products. They use only optimized firmware and update that firmware on a regular basis. The OCZ products we have tested today offer impeccable performance, yet OCZ is not the only leader among SSD makers who use the SandForce architecture.
It is the OCZ Vertex 3 Max IOPS and Kingston HyperX that deliver the highest performance among our today’s testing participants. The former is faster in more tests but the latter is somewhat less expensive, has an eye-catching exterior and comes with more accessories. Therefore we’d like to give our Editor’s Choice to these two.
On the other hand, the price of the leading products is far from affordable, so you may want to consider less expensive alternatives. If so, we’d recommend you to look at the OCZ Agility 3 and Patriot Pyro. They can help you save some money without losing much in terms of performance. These two have definitely deserved our Recommended Buy title.
The OCZ Agility 3 is currently a little cheaper on the American market, but the prices are constantly changing, so the Patriot Pyro may prove to be a better option after a while.