HD Tach 3.01
MTRON 32GB MSP 7000
Western Digital Raptor 150GB
Our HD Tach test results heavily favor the MTRON unit in all aspects except the burst rates. The MTRON's sustained transfer rate of 108.5MB/sec is about 43% better than the WD Raptor drive at 75.5MB/sec. The biggest difference is the MTRON drive holds the 108MB/sec transfer rate across the entire drive while the Raptor progressively decreases to around 52MB/sec at the end of the drive. The previous MTRON STR results were 95.1MB/sec, which is about 13% slower than the new drive. The burst rates of the Raptor at 128.8MB/sec are slightly higher than the 115.7MB/sec rates on the MTRON drive, but burst rates are not nearly as large of a factor as other indicators. Access times greatly favors the MTRON drive at .1ms compared to 8.4ms on the Raptor drive.
HD Tune 2.54
MTRON 32GB MSP 7000
Western Digital Raptor 150GB
The difference is slightly lower overall with HD Tune versus HD Tach, but the story remains the same. In our HD Tune test, the Raptor's sustained transfer rate of 72.1MB/sec is around 30% lower than the MTRON MSP 7000 SSD drive at 102.8MB/sec. More importantly is the fact that the MTRON drive again holds the 102MB/sec transfer rate across the entire drive while the Raptor progressively decreases to 50.4MB/sec at the end of the drive. The previous MTRON STR results were 90.8MB/sec, which is about 12% slower than the new drive. The burst rates of the Raptor at 100.6MB/sec are about 12% better compared to the MTRON unit. Access time greatly favors the MTRON.
Facebook
Twitter
RSS
48 Comments
View All Comments
alantay - Tuesday, December 4, 2007 - link
I'm hoping to see FusionIO's offer early next year. They target a $30/GB price and offer pci cards up to 640 GB and 700MB/s reads and 600MB/s writes. It seems like this is the technology that will matter in 2-3 years, if they get the price down (to $5/Gb maybe).http://www.fusionio.com/">http://www.fusionio.com/
mckirkus - Monday, December 3, 2007 - link
There is a good reason why they haven't released an IRAM 2.0. It would completely destroy the performance hard drive market and 150GB Raptors are cash cows.Think about it. With a SATA2 interface you would get well over 200MB/s sustained write and read and random read/write would be a non issue. Plus RAM doesn't wear out like flash.
And with RAM approaching $20 a gig you could have the fasted 32Gig HDD ever created for well under $1000. I would like to see Anadtech do a review of the HyperDrive4 as well. Gigabyte probably gets a cut of the revenue from WDs Raptors as long as they agree not to release an IRam 2.0.
That's not a conspiracy, it's just business.
mckirkus - Monday, December 3, 2007 - link
Did some digging. This thing at 32GB is $37 a gig.RAM is going for $19 a gig on NewEgg
http://www.newegg.com/Product/Product.aspx?Item=N8...">http://www.newegg.com/Product/Product.aspx?Item=N8...
Granted you need a device to hold the RAM and some battery backup but the performance is not even in the same league and the cost is about half of flash. An external SATA2 RAM drive with 32GB of 2GB RAM sticks would cost $600 for the RAM and more for the device. Why doesn't this thing exist? Any theories?
JarredWalton - Tuesday, December 4, 2007 - link
Your conspiracy would work, except there are numerous other problems with the iRAM. For example, the fact that you need to save the data to permanent storage before shutting down, and you need to initialize the data stored on the iRAM before it can be useful. Even then, most users rarely encounter situations where they are truly storage speed limited.That game you only get 12FPS running almost certainly loads all of the necessary data for a level into RAM if you have a 2GB machine, so the fastest storage system in the world is only going to be a bit faster. Even level load times might not improve much, as frequently loading levels involves a lot of additional logic like object instantiation, data parsing, and decompression routines.
The ideal approach would be a setup where you have a huge RAM cache that sits between the storage subsystem and the mass storage. Put some smarts on the cache so that it tries to keep the most useful stuff in memory, give it a battery backup, and commit writes to the actual drives as soon as it is reasonable to do so. That's basically how a lot of SCSI controllers work (and other top-end NAS type configurations).
iRAM was an interesting idea that really needs more work and tuning, and it's still quite expensive. Why limit transfers to SATA when even the slowest current RAM can sustain probably 10 times that data rate? Even better, get a 64-bit OS, figure out a way to add oodles of system RAM, and hopefully the OS can intelligently handle caching. *Hopefully!*
ChronoReverse - Tuesday, December 4, 2007 - link
Eh? The i-RAM has a rechargeable battery pack. DDR2 uses even less power than DDR1 so that pack will last even longer.Furthermore, if you simply Shut Down your computer, the battery pack isn't even needed. The i-RAM is kept alive with the standby power. The battery pack is only used when you unplug the computer completely or take out the i-RAM for which it'll last about 10 hours.
Ocire - Monday, December 3, 2007 - link
Hi,while crawling the net I found some interesting offers from Hama (a german company who mainly does Flash Memory etc.):
They sell a 64GB SSD (3.5", SATA) with a claimed 120/90 read/write for less than 1200 € (1700 US$) and the 32GB version for less than 700 € (1000 US$).
Are those relabeled Mtrons or is it their own product? How is their benchmark performance?
Here links to Hama and a german store who sells them:
http://www.hama.de/portal/articleId">http://www.hama.de/portal/articleId*159787/action*2563 (you can choose the language at the left, right above the customer login box, a little tricky to find ;-))
http://www.alternate.de/html/product/details.html?...">http://www.alternate.de/html/product/de...articleI... (unfortunately they don't seem to have an english version, maybe google or babelfish could either help or confuse you more ^^)
sparkuss - Monday, December 3, 2007 - link
Are you planning on adding any of the RAM drives to the equation? Such as the HyperOS HyperDrive 4. Or are they in a different class/economy to compete here?Reflex - Monday, December 3, 2007 - link
A couple years ago I worked with flash devices extensively, and it was fairly easy to run out of write operations. We could kill a SSD in about a week of stress testing, which translated by our estimates to about a year or two of typical desktop use. Thats not a very good lifespan for a $1200 product with so little storage space. Furthermore, in most cases when the writes ceased you could no longer read data from the device, which is even worse since there is no real warning.I'd really like someone to develop a true stress test for these types of devices. I would want a minimum of five years, with a warning when my writes are nearly finished.
PandaBear - Monday, December 3, 2007 - link
It is common to all NAND flash that the maximum amount of erase/program is about 100k for SLC, and down to as low as 500 for MLC (yes, 500, not 500k).Wear leveling is the key to the survival of these drive. Enterprises that use them for highly randomized read would have the best result. On the other hand, it is not really much of an advantage compare to a well designed system that runs on DRAM with battery backup and automatic hard drive fail over. I don't think I have seen these products personally yet, but I suspect it would be much better performance wise, but more expensive since you need to add a real HD and all the fail safe logics.
As we go into 50nm and 40nm process, the life will just get worse although the price get lower. I wouldn't be surprised if one day MS will provide a utility in windows that shrink your flash drive's capacity to get more life out of it. Or even better, provide a non-LBA based file system that reduce a whole lot of meta data / FAT table updates that slow down write and waste write cycles.
MrPickins - Tuesday, December 4, 2007 - link
Where did you come up with a figure of 500? All the documentation I see shows a typical endurance of 10k read/write cycles for MLC NAND.