Test Setup And Benchmark Configuration
|CPU||Intel Core i7-2700 (Sandy Bridge), 3.50 GHz, 8 MB Shared L3 Cache, Hyper-Threading enabled, Power-saving features disabled|
|Motherboard||Asus P8Z68-V, Z68 Express Chipset, LGA 1155, BIOS 3402|
|Memory||Corsair Vengeance 4 x 4GB DDR3-1600|
|Graphics||AMD Radeon HD 6970 2 GB|
|Storage||Plextor M5S 256 GB (PX-256M5S), SATA 6Gb/s, Firmware 1.00|
|Operating System||Windows 7 Ultimate (64-Bit) SP1|
|Intel Chipset Drivers||10.8.0.1003|
|Graphics Drivers||Catalyst 12.6|
|Anvil's Storage Utility (ASU)||RC3|
|hIOmon||Client Version 188.8.131.52|
|HD Tune Pro||5|
Accurately comparing SSDs based on different architectures is becoming increasingly difficult as more information about the behavior of solid-state storage comes to light. The preconditioned state of the drive, the workload, the length of time the workload runs, the randomness (compressibility) of the workload, the DRAM buffer, the internal algorithms, and OCZ's recent introduction of two MLC NAND modes all play different roles in defining performance at any given time.
In the interest of testing fairly and delivering reliable measurements, we're performing tests that evaluate the performance of brand new to positively dirty drives. Some folks only want to benchmark their hardware to make sure it matches the manufacturer specifications. The first part of our suite, therefore, employs Anvil's Storage Utility on fresh-out-of-the-box SSDs.
Drives typically incur the most intensive write activity during their first few days of operation as you install an operating system and applications. After that, the write activity of average client applications is estimated at around 10 GB per day. The second part of our benchmark is therefore based on real-world tasks that represent typical activities a client-oriented SSD might encounter within its first few days of use, and with a volume that does not yet have a whole lot of dead data scattered about its sectors.
The third component of our benchmark suite involves a check to see how write performance is affected when data is written across the full span using HD Tune. And the last part of our testing employs Iometer to explore garbage collection, along with the read/write performance of typical access patterns.
We wanted to make sure that we were comparing the M5S using the right stick. So, we selected SSDs marketed toward a similar user, priced similarly. Crucial's m4 utilizes the same controller and the same NAND as Plextor's M5S, but it only has a 128 MB data buffer. The m4's specified write performance is a lot slower than the M5S. However, Crucial's drive is really optimized for random I/O and not sequential throughput. Samsung's 830 uses an in-house controller and faster Toggle-mode NAND, but it only has a 256 MB buffer. Its faster flash is therefore offset by less of a cache, which is something we wanted to test for.
If you're wondering about the missing Vertex 4, we'll pit that against Plextor's M5 Pro in an upcoming story. Both drives exploit a newer Marvell 88SS9187 processor, which should make for an interesting comparison.
|Plextor M5S (PX-256M5S)||Samsung 830 256 GB (MZ-7PC256N)||Crucial m4 256 GB (CT256M4SSD2)|
|Controller||Marvell 88SS9174-BLD2||Samsung S4LJ204X01-Y040||Marvell 88SS9174-BLD2|
|DDR RAM Buffer Size||512 MB (256 x 2)||256 MB (128 x 2)||128 MB|
|NAND||Micron 25 nm Synchronous (29F128G08CFAAB)||2x nm Toshiba Toggle-mode DDR (K9PFGY8U7A-HCK0)||Micron 25 nm Synchronous (29F128G08CFAAB)|
|Formatted Capacity||238 GiB||238 GiB||238 GiB|
|Interface||SATA 6Gb/s||SATA 6Gb/s||SATA 6Gb/s|
|Form Factor||2.5", 9 mm||2.5", 7 mm||2.5", 9 mm|
|Warranty||Three Years||Three Years||Three Years|
|Sequential Read (MB/s)||520||520||500|
|Sequential Write MB/s)||390||400||260|
|4 KB Random Read (IOPS)||73 000||80 000||45 000|
|4 KB Random Write (IOPS)||70 000||30 000||50 000|