After the first generation of the SandForce controller, the successor was hotly anticipated. Although the data compression principle was a bit controversial at first, the second generation of SandForce controller logic launched an industry-wide trend in the SSD market. You could hardly find a single SSD manufacturer that didn’t have at least a few models based on the SF-2281 chip in their portfolio. The relatively accessible price of controller chips, which based their top-grade performance on the data compression principle, carried a few compromises along as well.
Firstly, the second-gen SandForce controllers didn’t have their own buffer memory, but used a part of the flash memory instead. This is why 64 GB SSDs actually came with 60 GB of usable space, and the same applies to 128 (120) GB and 256 (240) GB models. The second and more important reason was worse performance with incompressible data. Although most of the data a typical user works with daily is compressible, there are those who didn’t find this feature very attractive, due to the specific nature of their activities.
Goodbye SandForce, welcome Indilinx
Regardless of the abovementioned flaws, no one can possibly negate the fact that most of the current SSDs are based on this very popular controller. Even OCZ couldn’t escape it in their previous SSD generation (Vertex 3, Agility 3 etc.) and had to base those models on the said controller.
Reduntant NAND Array (RNA)However, OCZ didn’t hide their dissatisfaction with the controller’s performance as time went by, and the solution was found in the new, back-to-the-roots SSD generation, based on the Indilinx controller. Since we got the latest model, named Vertex 4, for testing purposes, let us remind you that the first Vertex generation had Indilinx’ Barefoot controller (SATA II, ~280 MB/s, 3K 4K IOPS). Vertex 2 (SandForce 1200, SATA II, ~285 MB/s, 50K 4K IOPS) and Vertex 3 (SandForce 2200, SATA III, 550 MB/s, 60K 4K IOPS) followed suit. Vertex 4 hearkens back to the beginning with Indilinx’ Everest 2 controller and improved performance, especially in the input/output operations segment (85K 4K IOPS). The use of the Everest 2 controller is no surprise, since OCZ acquired Indilinx Co. Ltd. around a year and a half ago. After the first-gen Everest controller, used to support OCZ’s Octane SSDs, Vertex 4 has arrived, with the second generation of Everest.
RNA is the fourth mechanism contained in the NDurance 2.0 technology, which, unfortunately, hasn’t been implemented in Vertex 4. It’s very important for data safety, as the mechanism works similarly to the one used in RAID arrays. Everest 2 automatically generates doubled data for each block, and then writes it to multiple NAND blocks. This gives the ability to retain data integrity even if a particular block goes bust. Errors occur, and they’re normally kept in check by the ECC, but this new ability makes the controller an additional backup layer in the case that ECC can’t cope with the given data. Although it hasn’t been activated in Vertex 4, RNA is a mechanism we can definitely expect in more expensive (enterprise) models, such as Intrepid 3, showcased at this year’s Computex.
The Octane series announced a major showdown between the resurrected Indilinx and the omnipresent Intel’s SandForce controller, and Vertex 4 starts it. Although many were expecting the SSD at hand to be named Octane 2, OCZ have decided to use its most well-known SSD model name to present Everest 2 to the masses. Although the time gap between the two generations of Everest was surprisingly small, OCZ claims that the differences between them are huge, and that virtually all shortcomings of the first generation have now been amended (mostly in terms of write performance).
Indilinx Everest 2
Adjustable NAND flash memory management (ANM)With regard to the fact that it’s made in 65 nm, the Everest 2 SOC isn’t all that different to the previous generation, at least not on the larger levels of work principle abstraction. The dual-core ARM chip’s clock has been raised from 333 to 400 MHz, which is the largest immediate improvement. Amendments have been made to the ECC engine itself too, as the latter is now capable of correcting up to 128 random bits for every kilobyte of data, unlike the first-gen Everest’s 78. For the sake of comparison, let’s just say that SandForce SF-2281 controller’s engine is able to correct 55 bits per 512 bytes of data (which amounts to 110 bytes per KB). The most important hardware improvement has taken place on a somewhat lower level, and that’s the indifference of performance to data compressibility, i.e. equally good performance regardless of whether particular data has a high or low degree of compression. Furthermore, the firmware has been improved as well – the firmware balances between work modes depending on whether the disk is less than 60% full or not. This ensures optimal work with both small and large files, enabling consistently good performance.
ANM is yet another mechanism of NDurance 2.0 technology, in charge of maintaining an SSD’s lifetime and performance. It’s basically the controller’s capability to apply both the manufacturer’s commands and its own. Actual examples would be internal changes to cell voltage and sophisticated signal processing techniques during the data read/write/erase cycle. Lowering the cell voltage gives them a clearer signal, especially during stressful processes, which reduces the physical deterioration of NAND flash memory, thereby increasing its lifetime.
Write amplification (WA) is a phenomenon typical of flash memory, and by extension SSD drives. It’s basically an unwanted effect that happens during data write, where the physical amount of information that’s supposed to be written gets multiplied by the logical quantity required for it to be written. We’ll clarify. In order for data to be written, the flash memory has to be erased first. This process involves moving or writing data and its metadata more than once. This multiplication effect increases the number of writes, which reduces the lifetime of SSDs, i.e. their cells, due to the limited number of writes they can handle. With data compression, the WA reduces significantly. Factors that impact WA directly are garbage collection (the algorithm’s efficiency in choosing the best block for data rewrite), over-provisioning (the percentage of the physical SSD capacity allocated by the controller, which the user cannot access), TRIM (SATA command sent by the OS to tell the SSD which data can be ignored during the work of garbage collector), free space available to the user, safe erasing (resetting the SSD to its factory state), separating static from dynamic data (grouping data according to the tendency to change), sequential data write and random data write.Besides the fact that it supports the SATA III interface, Everest 2 also has the NDurance 2.0 technology, automatic 256-bit AES encryption, the said ECC engine and flexibility towards different versions of NAND flash memory and commands specific to particular manufacturers of NAND chips. Below the mask of the SSD itself, you’ll see that the PCB of the new Vertex 4 is projected so that the controller is centrally mounted, with chips surrounding it. The chips in question are Intel’s 25 nm NAND memory chips, which, having in mind the price and performance of Intel’s chips, seems like a logical move. The controller is surrounded with 16 chips of 8 GB, eight on either side of the PCB, while caching was assigned to 512 MB of Micron’s DDR3 800 MHz memory, in two chips of 256 MB (one on each side).
NDurance 2.0Another new technology brought forth by Everest 2 is NDurance 2.0. It entails certain mechanisms that help maintain a longer lifetime for memory chips, and thus the drive on the whole. With NDurance 2.0, OCZ has upped the limit for standard warranty to five years, which is very important and says a lot about product quality. The four most important aspects of NDurance 2.0 are: reduction in the write amplification of compressionless data (WA), an advanced ECC engine, adjustable NAND flash memory management (ANM) and RNA. All of these are explained in detail in separate boxes, for those who wish to delve a bit more in-depth into the matter at hand.
PerformanceBefore shifting the focus to real-world performance, we’ll have a closer look at the features that make Vertex 4 more advanced than any previous SSD according to OCZ. Firstly, it’s a drive with IOPS performance exceeding 120.000, while the lack of any impact of compressibility on performance is stated as a separate benefit. Furthermore, QD 1-16 performance, which always reflects the realistic power of an SSD, is constant, yielding good results even with the minimal multitasking of portable PCs. Finally, the number of IOPS isn’t limited by write cycles. Performance measurement consisted of the standard testing battery with five popular tools: ATTO Bench, AS SSD, CrystalDiskMark, SiSoft Sandra 2011 and PCMark 7. The idea was to measure exact read and write speeds, and we picked a Verbatim model for comparison, as it had the same capacity, but was based on SandForce 2281.
|Test Results||OCZ Vertex 4 128GB
||Verbatim 120 GB
||Kingston HyperX 240GB
|ATTO Bench 2.41|
|Read QD4 0,5 / 4 / 8 / 265 / 8192 [KB/s]||15.028 / 95.067 / 144.457 / 455.658 / 528.936||14.592 / 109.092 / 180.419 / 483.339 / 538.066||21.651 / 166.569 / 250.478 / 536.665 / 560.538|
|Read QD10 0,5 / 4 / 8 / 256 / 8192 [KB/s]||17.920 / 129.249 / 218.665 / 465.476 / 463.819||34.984 / 274.916 / 370.360 / 529.998 / 479.349||39.808 / 293.011 / 401.080 / 557.784 / 650.538|
|Write QD4 0,5 / 4 / 8 / 256 / 8192 KB [KB/s]||56.562 / 314.752 / 379.651 / 382.892 / 389.978||14.356 / 251.896 / 332.309 / 517.196 / 522.502||18.560 / 227.950 / 347.832 / 524.802 / 526.344|
|Write QD10 0,5 / 4 / 8 / 256 / 8192 KB [KB/s]||55.552 / 297.107 / 383.736 / 382.892 / 376.311||28.800 / 342.712 / 420.035 / 515.949 / 423.960||30.515 / 366.264 / 450.678 / 524.802 / 522.926|
|AS SSD Benchmark 1.6|
|Read Seq / 4K / 4K-64 Thrd [MB/s]||489,1 / 24.7 / 333.3||204.1 / 18.2 / 95.5||518.8 / 22.9 / 188.2|
|Write Seq / 4K / 4K-64 Thrd [MB/s]||341.9 / 93.1 / 289.4||83.5 / 61.7 / 47.1||239.2 / 93.2 / 212.4|
|Read 16 MB / 4K / 4K-64 Thrd / 512 B [IOPS]
||30.57 / 6.329 / 85.317 / 7.254||12.75 / 4.646 / 24.454 / 6.335||32.4 / 5.868 / 48.183 / 7.676|
|Write 16 MB / 4K / 4K-64 Thrd / 512 B [IOPS]||21.37 / 23.822 / 74.079 / 38.827||5.22 / 15.800 / 12.062 / 3.679||15 / 23.855 / 54.380 / 4.911|
|Copy Benchmark ISO / Program / Game [MB/s]||240.9 / 158.3 / 229.6||107.2 / 53.9 / 86.2||164.2 / 153.6 / 164.4|
|CrystalDiskMark 3.01 x64 Incompressable data (random)
|Read Seq / 512K / 4K / 4K QD 32 [MB/s]||444.4 / 336.7 / 29.44 / 358.5||208.2 / 199.2 / 28.3 / 94.8||522.1 / 462 / 37.2 / 211.9|
|Write Seq / 512K / 4K / 4K QD 32 [MB/s]||368.1 / 382.4 / 103.9 / 331.9||147.5 / 147.8 / 78.3 / 143.9||225.6 / 226.9 / 110.4 / 226|
|CrystalDiskMark 3.01 x64 Compressable data (All 0x00, 0Fill)
|Read Seq / 512K / 4K / 4K QD 32 [MB/s]||427.1 / 346 / 29.03 / 350.1||472.2 / 442 / 33.71 / 117.8||506.6 / 457.4 / 38.06 / 201.6|
|Write Seq / 512K / 4K / 4K QD 32 [MB/s]||387.7 / 402.5 / 104.3 / 329.5||490.4 / 482.6 / 81.25 / 354.5||492.4 / 469 / 103.1 / 317.9|
|SiSoftware Sandra Pro Business 2011.SP5a|
|Read / Write [MB/s]||517.18 / 369.1||540.15 / 81||533.8 / 297.4|
|Random Access Time [ms]||7||24||2|
|Test Platform: Intel Core i7 3960X, ASUS Rampage IV Extreme X79 (Intel SATA 6 Gb/s), 4x 4 GB Kingston HyperX DDR3 1600 MHz, AMD Radeon HD7850, CM 900W, Windows 7 Ultimate 64-bit|
Waste collectionATTO Bench barely finished its run, and it was already clear that write rates for small files were considerably better in Vertex 4 than its SandForce competitor, even with QD 4/10, and without much oscillation between 8 KB and 8 MB, which is really rare. With the QD 4 setting, read speed increases together with file size, while QD 10 reaches full performance at 64 KB already. The queue depth (QD) setting reflects the behaviour of the drive under different loads and can be compared quite directly with drive access requests. The smaller the QD, the faster the drive’s response, which means that a higher QD will emulate a large number of applications accessing the SSD at the same time, which is a scenario that Vertex 4 handles excellently, both in read and write aspects.
You can freely look upon Garbage collector as your SSD’s local waste collection service, utterly needed due to the “rubbish” your SSD makes when writing data. Data is written to an SSD in pages, comprised of individual cells. Pages are then grouped into blocks. Each block is written into as long as there’s free space, but page erasing can’t be done individually, and the entire block has to go through the deletion process. The process of data writing looks like this. A block of twelve pages contains data in four pages (A-D), while the other eight is empty. Four more pages of data are added into the block (E-H), but data is also changed for the previously written four pages (A-D), which are then pronounced by the controller to be stale pages. The changed data is then written into the remaining four empty pages (A’-D’). In order to use the space inside the block that contains the unnecessary stale pages, all good pages (E-H and A’-D’) are copied into a new block, while their own block is erased and becomes available for a new write. This block-erasing step is called garbage collection.
Sequential read/write leaves the Verbatim model in the dust. The same applies to the I/O operations test, where OCZ introduced a really big performance hike. The difference is the most evident in small file manipulation, where results have been better up to ten times! Furthermore, the scenario that’s supposed to reflect real-world situations of copying files of various types and sizes (ISO, programs, games) is overbearingly in favour of Vertex 4, with results being twice, even three times better. The application that was supposed to show the widest gap between SandForce- and Everest-based controllers is AS SSD, a test with uncompressible data. Expectedly enough, sequential read/write tests are largely in favour of Vertex 4. The competitors remain neck-to-neck even with compressible data, with Vertex 4 handling higher QD values better. Finally, after similar results were yielded by Sandra, PCMark 7 also showed a huge advantage of Vertex 4 over Verbatim, mostly due to better IOPS performance.For the sake of comparison, although it belongs to a different price category, we compared Vertex 4 with Kingston’s HyperX 240 GB, the latter being Kingston’s strongest SSD model. Although it has a higher capacity, making its performance slightly better than other versions, it was the perfect way of comparing Everest 2 with the shiniest example of SandForce 2281. As was the case with the Verbatim model, although the maximum sequential read/write speeds reach higher than the Vertex 4 we had on test (the weakest one in the series), small file manipulation is still multiple times better than anything Kingston could produce. The situation with IOPS is similar, and the same goes for real-world copying test in AS SSD. Expectedly, it was an even field with compressible data, small files and a high QD already lean towards Vertex 4, while uncompressible data shows its full sovereignty.
A fitting name
Although we were a bit sceptical towards the performance promised by the new Indilinx controller, the test results dispelled any possible doubt. A finely balanced controller with phenomenal NAND cell lifetime preservation mechanisms speaks volumes about OCZ’s step in the right direction. Constant performance levels, excellent small file manipulation, very similar behaviour under different loads and fantastic IOPS performance are all disciplines in which Vertex 4 sets new limits. The problems competitors faced have been avoided for the largest part, and perhaps the best thing about the new Vertex 4 product line is the NDurance 2.0 technology and the 5-year warranty, which clearly speaks that it’s not just about speed, it’s also about reliability with this latest newcomer to the market – OCZ stands behind their claim that no performance degradation will occur in their drives for at least five years, and that’s a lot of time.
|OCZ Vertex 4 128 GB
|Controler||Indilinx Everest 2|
|Interface||SATA 6 Gb/s|
This isn’t the fastest, the cheapest or the largest drive on the market, but it’s definitely the optimal choice all things considered. We’re expecting even better performance from higher-capacity models, but we loved the 128 GB version very much. Although OCZ still has SandForce-based models in their portfolio, it’s about time for the mid-range to say goodbye to SandForce and welcome to Indilinx.