FOREWORD
I would like to say a few things before I start this review. First, it is my first hardware review. I've played with a lot of hardware, overclocked stuff and been interested in high-speed storage for a long time. I used to run a 15000rpm scsi setup for years(until going sff) and was one of the first to get my hands on a core ssd when ocz came out with them. After some problems with that, I resorted to running a ramdsik to mirror my small boot partition to ram after boot. Lots of messing around, but no documentation or review, so I'll do as well as I can.
Secondly, this will not be a conventional review. The acard will lose in every benchmark, as I'm comparing it to a software ramdisk. It's both a fair and unfair competition in many ways. Fair in the sense that the storage medium is the same, and that both will have some of the same problems (size, data lifetime) and unfair in that system ram has no real bottleneck, and is in many ways, the fastest disk that you can possibly get. The point of this review will be to see how large of a difference there is for practical purposes, between the acard ramdisk and the software ramdisk. A review will normally have the reviewed product compare favorably with it's competition, but the purpose here is not to show such. It is not to compare it against rivals, because there are no competitors in it's class. From the Acard 910, the only step up is another Acard 910 or using a software ramdisk.
Some may argue that the competition for this drive is a high end flash-ssd, but I see the two as completely different products for completely different uses. Yes, both are fast and have no moving parts, but the similarity ends there.
Test setup:
Q6600@3.6ghz 1.45v
8gb ocz ddr2-1000
EVGA GTX280@670/1404/1150 stock cooling
520HX corsair power supply ----- half-size p35 shuttle!!
Viewsonic VP2290B 3840*2400 (22.2") ----- 30" 3007wfp-hc 2560*1600 lcd
And of course, disks:
1TB hitachi 0a38016 32mb cashe 7200rpm disk
6gb mirrored partition with supervolume (6gb ramdisk using system ram)
16gb ddr2 + acard 9010 sata ramdisk in various configurations
SATA controller: intel ich9r in AHCI mode, pcie frequency=100MHz
SOFTWARE RAMDISKS
Software ramdisks are not commonly used by the overclocking community, and I'm not sure why they never caught on. They allocate a large chunk of system memory and install a virtual disk driver, which allows you to access that ram as if it were an additional hard disk. The two that are used to compare behave slightly differently. Supervolume takes an existing partition and loads the entire partition to memory. All accesses to the partition go through ram first. Reads are read straight from ram, writes are first written to ram, then while the disk is idle, the data is slowly written to disk in the background. Superspeed ramdisk creates an empty drive which can be saved and loaded from an image file. Both have the effect of data persistence, but use space on a hard drive to achieve this. Also, you can't truly boot from either, though supervolume gives the same effect after it loads the data, but this adds significantly to the boot time. They also obviously eat up your system memory and supervolume limits your boot partition size, which can cause some applications to complain.
INTRODUCTION
There are a lot of options in the storage department, but a few stand out, the ACARD 910 being one of them. An interesting little device, it turns up to 8 sticks of ddr2 into two sata drives. Let's start with a few pictures.
FEATURES
8 slots for ddr2. Anyone who buys this drive will fill it up to 16gb. While it can supposedly do 64gb, it is cheaper to get an additional ramdisk filled with 2gb sticks.
2 sata ports. Raid-mode gives additional sata bandwidth. No other drive currently on the market is actually limited by the sata interface, but this is a very useful features for this unique drive. The drive is seen as two physical disks at half of the original capacity, ready for raid.
LI-ION battery backup. A battery keeps the ram from losing data while the computer is off. A pair of standard-size cells are used with a simple protection circuit.
Compactflash backup. Backs the drive up to and from compactflash. The slot is located at the front of the unit, so it is easily accessible when installed.
5" bay. The whole drive takes up a 5" bay with a mirrored smoke plexiglass front.
With the 9010, there are some options, set with jumpers on the back of the drive:
Jumper settings
SATA MODE: (150/300) for legacy systems I don't see this as much of a real option. It adds legacy support for a product that will only realistically be used with high-end hardware. I tested it anyway.
SATA PORT MODE: This jumper sets the option to split the drive into two logical drives, each with half capacity and nearly full-performance. This option was tested extensively.
ECC MODE: This setting was hidden under a "reserved" sticker and mentioned in the manual. It bsically toggles soft-ecc mode on and off. ON gives you soft-ecc at the expense of space, off gives you the lost storage back.
BATTERY LIFE
With 16gb of ram (8x2GB), I charged the battery to full by leaving the ramdisk plugged in overnight. I then unplugged the ramdisk and monitored the power led inside the drive. The ram lost power at the 14hour mark, so it's good enough to take on a long flight. However, this will cause incompatibilities with people's schedules. I don't like to leave my computer on if I'm not home, and I know others feel the same way. Acard has three options here: use standby power from the computer, use a bigger battery, or use a separate power supply. I know it has compactflash backup, but that takes time too, and would be a hassle for people to do daily. Battery life would ideally be a bit over 24hours to accommodate for variations in users' schedules.
SYNTHETICS
Performance: How do the above options affect performance of the 9010? Synthetic tests show various configurations below. The notepad window near the top contains the test setup.
We can see that the results are very consistent. ECC and the number of sticks do not affect performance, and dual-sata mode performs equally well on each sata interface, with individual port performance on par with the single-sata mode. We can also see a large difference between software raid (server 2003 striped dynamic volume with default settings) and hardware raid (intel ich9r with default settings). I tested software raid because I expected it to be fast. It basically gets around controller limitations and allows the true potential of the drive to be tested. If run on a very high-end sata raid card, I'd expect the results to be similar to what is seen with software raid. As I don't have a very high-end sata raid card, the software raid numbers are the best I can achieve.
For the sake of comparison, here are the benchmarks above, run on the ramdisk:
Tough competition, but that's what you'd expect from system ram. Note the odd ATTO results and the lack of hdtune and hdtach. Some benchmark programs are not software-ramdisk friendly and give odd results or don't detect the drive.
So, let's do a few tests that reflect what you might actually do with the drive:
9010 mode: dual-sata mode, ecc off, 16gb size, software striped logical volume
Notes about the image: All time is measured in seconds (all results are timed). Some of the tests were timed manually and have some error (only winrar is machine-timed)
About the tests:
FILE TESTS:
7zip game: A 7zip compressed game was installed for the 7zip test, installing+extracting at the same time, with lots of file operations, but largely cpu-limited as noticed in the task manager.
RAR ISO: Using default winrar settings, the original starcraft iso was compressed.
UNRAR ISO: Using default winrar settings, the rar generated from the RAR ISO test was extracted.
IMG RAR: Using default winrar settings,a folder containing a few hundred images and a few small videos (1gb total) was compressed.
IMG UNRAR: Using default winrar settings, the iso from above was decompressed.
IMG RESIZE: The folder of test images was resized to 1024*768, leaving original images intact (additional files created).
COPY TESTS:
ISO COPY: The starcraft iso was copy-pasted into the same directory, time measured.
IMG COPY: The test image folder was copy-pasted into the same directory, time measured.
3GB selfimage: A 3gb image was created and written to the same drive (ramdisk and acard used partition->partition copies, hard drive used image->image copy)
BOOT/APP TESTS:
CLEAN BOOT: Clean boot time was measured within vmware.
ATTO BOOT: Clean boot time was measured within vmware while atto bench was testing.
BUSY BOOT: An "old" windows install was simulated by loading a lot of programs at startup. Boot time measured from hitting enter at boot selection menu to final application loading completed.
APP LOAD: The same programs loaded for the "busy boot" test, but measured without the windows boot time. To show the difference in actual application launch times.
Note that vmware is not perfect and there is a performance hit for ALL drives, a native setup will be significantly faster, but relative (in-vmware) results should be consistent.
CONCLUSION
It's fast, very fast. Don't let the “dismal defeat” look of the graphs sway your opinion, it's compared to system ram, not something you normally use for storage. In many of the tests, it was able to at least keep up with the system-ram based ramdisk, which is impressive. Also, it doesn't eat your system memory, limit your boot partition size, increase boot time, or get wiped immediately at shutdown (as a system crash will do to a software ramdisk). However, it is not perfect, and I have made a list of improvements that I'd like to see with the second version of the product.
As a side-note, I took it to a lanparty over the weekend, where i used it to store+serve the games we were playing. Everybody was impressed with how fast the maps loaded.
Things to note:
The amount of ram installed does not change the performance of the ramdisk.
ECC does not change performance either.
You will need a good sata controller to get good raid results.
IMPROVEMENTS
Use notebook dimms. Smaller, lower-power, and cheaper. At the time of reviewing, 2gb sticks were available for only $16 on newegg without any rebates. You can easily fit 16 or more laptop sticks in the same space, or even in a 3” bay.
Use a li-polymer cell the size of the pcb mounted on the top of the casing that is not over memory slots. Maximum use of space.
Switch interfaces. Use Pcie-16x or include many more sata interfaces. The drive is severely sata-limited.
The per-sata bandwidth is slower than what can be done. Use code tweaks to get it up to par. Sata-2 can do more than 170mb/s, might as well use it. 250+mb/s through 8 ports is a matter of adding a few more sata PHY chips and updating the firmware a bit.
The unit needs 24 hours of battery life. People run on a 24hour clock, so having the battery life being shorter than 24hours can require the computer to be left on continuously rather than on-demand. For example, if I went to sleep at midnight and woke up at 8, went to school/work until 5pm, the drive would be erased if the computer was off. Unfortunately, that's a very common daily schedule. consider a 2s2p battery pack.
Reply With Quote
Bookmarks