1. About the OCZ RevoDrive 50GB
We have in our labs the OCZ Revodrive, a high performance PCIe Solid State Drive (SSD) for consumer applications in desktops and workstation PCs. Moving beyond the bottleneck of SATA II, the RevoDrive features a PCI-E interface and promises to deliver speeds up to 540MB/s reads and random 4k writes up to 75,000 IOPS.
The device has been designed for high-performance gaming PCs and workstations, allowing users to run both Windows and applications.
OCZ is offering the device in capacities of 50GB and up to 480GB, a 3-year warranty and a 2 million hour mean time between failure (MTBF) rating.
The idea behind the Revodrive is to use the PCIe interface of your system in order to connect your solid state storage medium, keeping SATA around as an interface with traditional hard drives, especially if your system is not supporting 6Gbps SATA. By putting NAND in parallel, the Revodrive offers a high-capacity and fast solution that could be used for both booting up your system and of course storing your applications.
Today we will run some benchmarks on the 50GB version of the drive (part number OCZSSDPX-1RVD0050), which can be purchased online for $200~230.
- Capacity: 50GB to 480GB capacities
- Part Number: OCZSSDPX-1RVD0050
- PCI-Express interface (x4)
- For use as primary boot drive or data storage
- Internal RAID 0
- 181.07 (L) x 21.59 (W) x 125.08mm (H)
- Shock Resistance: 1500g
- Seek Time: 0.1 ms
- Operating Temp: 0°C ~ 70°C
- Storage Temp: -45°C ~ 85°C
- Power Consumption: 3W Idle, 8W active
- MTBF: 2,000,000 hours
- 3-Year Warranty
- Compatible with Windows XP 32/64, Vista 32/64, Windows 7 32/64
- Read: Up to 540 MB/s
- Write: Up to 450 MB/s
- Sustained Write: Up to 350 MB/s
- Random Write 4KB (Aligned): 70,000 IOPS
2. A closer look
Below you see the retail box of the RevoDrive
In addition to the 50GB (part number OCZSSDPX-1RVD0050) drive you get the drivers and a manual:
OCZ has shipped the drive into appropriate and secure packaging:
The RevoDrive uses two SandForce SF-1200 controllers in RAID-0 on a PCIe x4 card.
Compared to a pair of SF-1200 based SSDs in RAID-0 connected to a motherboard’s RAID controller, the Revodrive will not offer you any different performance, although the CPU utilization of Revodrive's onboard Silicon Image RAID controller could be lower and of course, a standard RAID setup would cost more and would also add some extra steps to installation.
The Sandforce SF-1222 controllers have been installed in many SSD solutions, it features the , DuraClass technology for reliability, endurance and performance.
DuraWrite technology promises to extend the life of the SSD over conventional controllers, by optimizing writes to the flash memory and delivering a write amplification below 1, without complex DRAM caching requirements.
The SF-1200 has also built in AES-128 bit encryption controllable by a configurable user password.
A Silicon Image RAID controller and a Pericom PI7C9X130 PCI Express bridge chip are found on the PCB. The PI7C9X130 chip is obviously responsible for the PCIe 4x link.
The 50GB RevoDrive we have in our labs has 60GB FLASH NAND on there the PCB spread out over 16 ICs, 8 ICs on each side. As you see in the picture below, there is free space left for more NAND chips on the PCB of the specific model. The same PCB can accommodate more NAND chips and offer higher capacities:
The card has a x4 PCIe lane design but it still can be installed into a x16 slot. Below you see the funky activity LEDs of the installed card:
Installing the RevoDrive is simple. In case you plan to use the RevoDrive as secondary drive, boot into Windows and install the essential driver, which can be downloaded from OCZ's web site. After installation you go to disk management and setup/format the drive. After that it's ready for usage. Since the RAID controller is per-configured (Raid 0) , there is no need to setup anything in RAID. Here is a screen shot of the BIOS :
On the other hand, if you like to have your OS installed on the drive, then in the BIOS set it to boot drive. You need to supply Windows with the appropriate drivers to recognize the controller. When Windows ask you where to install the OS, point windows to the RAID driver on the CD. After installation, the RevoDrive partition will appear as root drive.
After installing the drive, the Silicon Image Sil 3124 SoftRaid 5 controller appears in Windows's Device Manager:
The firmware revision of out model is 1.20. Each part of the RAID drive is offering 25GB of capacity:
Here is our testbed:
- Motherboard: Asus Striker II Extreme Bios 1403 (Nvidia 790i Ultra SLI)
- Processor: Intel Q9300 (Quad Core @ 2.50GHz)
- Case: Open Air testbed
- Power Supply: OCZ GameXStream GXS600 SLI-Ready
- Memory: 2x1GB Supertalent DDR3-1600 (1333MHz@ 7-7-7-20-1T)
- HDD: Seagate 7200.12 500GB SATA2 7200RPM
- Monitor: LG L246WH-BH 24"
- Windows 7 32bit with all the latest updates installed
The SSD was not used as a boot drive, so we did not perform any kind of optimizations for the operating system. The "enable advanced performance" option was selected under the device's properties (OS). For the tests, we used the following software leaving their settings to defaults:
- HDTachRW v188.8.131.52
- HD Tune v4.60
- Crystal DiskMark v3
- ATTO Disk Benchmark v2.46
- ASS SSD Benchmark 1.5xxx
- IOMeter v2006.07.27 with Xtreme Benchmark template
We start the tests with the HDTachRW software. HD Tach is a low level hardware benchmark for random access read/write storage devices. The software measures the sequential read speed (at various points on the device), the random access speed and sequential write speed. We selected the full benchmark.
The software reported an average sequential writing speed of 229.9MB/s and a sequential reading of 232.1 MB/s. The results is slower than we expected , possibly due to the RAID controller of the RevoDrive. In addition, both the read and write performance graphs are not constant.
OCZ explains that "HDTach’s fixed strides between sample points causes issues. The software uses fixed strides across the array to measure small block sizes. The periodicities of the stride pattern and the flash memory address space may cause some wave-like
interference patterns of the benchmark results that are also measuring artifacts rather than being
indicative of performance. "
The HD Tune Pro software is also a utility we used to measure the drive's reading performance. Although not necessarily representative of real-world workloads, HD Tune's targeted tests give us a glimpse of each drive's raw capabilities.
Let's start with a sequential read test. The drive read the data at 205.6 MB/s (average), significantly slower than we expected. As you see below, the fist 13GB of the stored data was read really slowly.
Here is another sequential file test. We used 64MB of data comprising out of sizes ranging from 0,5 KB to 8 MB. As you can see, the RevoDrive showed its strength once it passed the 128KB file size. The performance for write topped at around 425MB/s and for read close to 500MB/s:
The HD Tune Pro also allows random access read testing. The 1MB performance topped at 361 MB/sec:
The next software we used was the ATTO Disk Benchmark. The tool measures storage systems performance with various transfer sizes and test lengths for reads and writes. The benchmark performs file transfers ranging from 0.5 KB to 8192 KB.
This time the drive gave the expected performance an t average of 450 MB/s for write and more than 500 MB/s for read , depending on the file size, in Queue Depth 4:
Generally, ATTO is the preferred Standard benchmarking software as while it runs spot data is has several advantages over other HDD benchmarks including the fact that it shows the specific speed of each file size transfer and shows a true average, this benchmarking product also doesn’t seem to favor SRAM over DRAM caches .
The next benchmark is the CrystalDiskMark. The software provides throughput data based on sequential reads and writes, and random (512K/4K/4KQD32) reads and writes. We've used the default 1,000MB file-size for the tests. We start with the worst case scenario, where we used the software's random fill pattern (default settings), so the results are not what we could expect from the RevoDrive. Let's say that this is what you'll get in the worst case scenario for the drive:
Sequential write is low while the 4K thread is very high.
We changed the fill pattern of the file size and ran the test again. As you see below, Revodrive's sequential performance has been restored to normal levels:
The AS SSD benchmark contains five synthetic as well as three practical tests.
The synthetic tests determine the sequential and the random read / write performance of an SSD. These tests are carried out without using the operating system's cache. The Seq-test measures how long it takes to read and write an 1GB file.
The 4K benchmark tests he read and write performance for random 4K blocks. The 4K-64-THRD-test corresponds to the 4K procedure except that here the read and write operations are distributed on 64 threads.
The drive scored 430 points at the AS SSD Benchmark. Although most of the performance figures you see below are lower than expected, the 4K file write performance is good:
Looking at the following test, we realize that the RevoDrive is very fast when if deals with fully compressed files. Reading maxed out at 466 MB/s and writing at 441MB/s, with 100% compressed data:
Finally, we proceed to IOMeter benchmark. Iometer is run by using workstation and database patterns for queue depths (outstanding I/Os) of two and 32, representing very light and moderate loads. Iometer is both a workload generator (that is, it performs I/O operations in order to stress the system) and a measurement tool (that is, it examines and records the performance of its I/O operations and their impact on the system). The app's ability to bombard drives with an escalating number of concurrent IO requests also does a nice job of simulating the sort of demanding multi-user environments that are common in enterprise applications. It can be used for measurement of the performance of an SSD. We run the IOMeter tests using the Xtreme Benchmark template .
Here is the test setup:
Below are the results:
The RevoDrive scored very high in the total I/O test and was also impressively fast with an average response time of just 0.529 ms.
As a concept, a PCIe based SSD is a smart way to enjoy the advantages of RAID0 without having to buy two separate SandForce-based SSDs and messing up with many installation settings, as long as it is reasonably priced. Especially if you are dealing with large and compressable files and you need your SSD to operate under really heavy loads, then RevoDrive is your solution.
Although this RAID0 array of two SandForce controllers hidden behind a PCI Express interface is able to give you as much as double the performance of a single SSD, this performance heavily depends on the the situation. According to out tests, the RevoDrive will not be as fast as you may expect in every and any scenario .Specifically, it seems that SandForce’s DuraWrite compression technology is relatively slow when dealing with incompressible data, such as is used in the AS SSD benchmark. This is the case with the Revo, as we saw it manage a sequential read speed of 319MB/sec and a sequential write speed of just 90.37MB/sec rather than the quoted 540MB/sec and 450MB/sec. This performance is close that of some single SSDs. Of course, the same test showed that when the drive accessed/wrote 100% compressed data, the performance was increased to meet the drive's specifications.
We also experienced some surprising low sequential read and write results in the full benchmark of the HDTachRW test. On the other hand, when the same benchmark used many smaller and thus more compressible files for the sequential test, the RevoDrive maxed out at around 425MB/s for write and close to 500MB/s for read, especially when it was accessing /writing files larger than 128KB.
In Iometer, which randomly reads and writes three consecutive patterns of 4KB files, the Revo was also not as fast as expected. In the reading part, compared to single SandForce-based SSDs, the RevoDrive was nearly twice faster. The performance during writing depended on the queue depths. In low queue depths, the random write performance was as low as what you may expect from a single SSD. But if you increase the queue depth for a particularly intensive workload, the the RevoDrive will take the lead. In addition, the drive showed impressively low response times.
Every benchmark uses different methods of measuring performance and some benchmarks appear
not to be able to measure SSD performance, while others are somewhat better.
With these results in mind, we believe that the majority of desktop users aren’t pushing this many IOs at a given time so for them, the 50GB RevoDrive could not be the best solution for them, compared to a pair of SandForce-based SSD, which will additionally offer the benefits of TRIM support. On the other hand, if you plan to add your SSD under heavy loads, then the RevoDrive is a good solution, provided its good price.