Sign in with
Sign up | Sign in

Benchmark Results: Streaming And 4K Reads/Writes

Super Talent's 512GB RAIDDrive: RAID On An Add-In Card
By

Here is the effective throughput of the RAIDDrive. It maintains 720 MB/s, even when utilizing most of the drive’s capacity. The peak throughput of 892 MB/s is actually higher than what we’ve seen in ioDrives or ioXtreme drives by Fusion-io. You need two ioXtreme drives to beat this number. Once again, we want to point out that Fusion-io ships capacities of 80GB, 160GB, and 320GB, while Super Talent can easily provide 512GB or 1TB capacities, depending on what you’re willing to spend. In any case, a cost per gigabyte comparison definitely falls in Super Talent’s favor.

The RAIDDrive also shines when it comes to streaming writes; ioXtreme drives just can’t compete here. Roughly 500 MB/s sequential, streaming write performance simply rocks.

SSDs can write 4KB blocks at a time. However, they typically have to delete much larger blocks (read-erase-modify-write), which can cause severe performance impacts. This is where Fusion-io shines with all of its products and settings, so long as you choose a configuration that optimizes for write performance. If you don’t, Super Talent actually performs rather well. Intel’s X25-M G2, which drops significantly in this test, offers a good comparison.

Display all 25 comments.
This thread is closed for comments
Top Comments
  • 14 Hide
    barmaley , January 6, 2010 5:18 AM
    Conclusion page reads "More importantly, though, the RAIDDrive is bootable, while Fusion-io products currently are not. Driver support is available for virtually all relevant operating systems."

    Did you even read the article ?
  • 10 Hide
    mitch074 , January 6, 2010 6:49 AM
    @IzzyCraft: it's just you. They didn't splice 2 SSDs.

    They spliced 4.
Other Comments
  • 14 Hide
    barmaley , January 6, 2010 5:18 AM
    Conclusion page reads "More importantly, though, the RAIDDrive is bootable, while Fusion-io products currently are not. Driver support is available for virtually all relevant operating systems."

    Did you even read the article ?
  • 3 Hide
    barmaley , January 6, 2010 5:52 AM
    Well, at current price points these drives are waaaay out of my reach. However, it is nice to see fast hardware like that to come out. And, of course, we all know a few years from now it will only cost a fraction of the price today.
  • 0 Hide
    IzzyCraft , January 6, 2010 5:58 AM
    Is it me or does the picture look like they spliced 2 ssd drives onto a pcb
  • 10 Hide
    mitch074 , January 6, 2010 6:49 AM
    @IzzyCraft: it's just you. They didn't splice 2 SSDs.

    They spliced 4.
  • 0 Hide
    cjl , January 6, 2010 7:08 AM
    hakestermanYes, I agree if you can't boot from them then their pretty much useless. However booting from them is just n a Bios upgrade away. But getting every board manufactuer to get on board with this may be tricky.

    BIOS update? Nope. The issue is drivers. You need to be able to load the appropriate drivers for the OS to see it.
  • 0 Hide
    g00ey , January 6, 2010 7:38 AM
    Regardin compatibility; I seriously doubt that this card is compatible with OpenSolaris which is a relevant operating system. Intel IOP333 works with OpenSolaris from what I've heard but the IOP348 is certainly not in their HCL. Highpoint's cards are known not to work and many of their RocketRAID cards are based on the IOP348 chip.
  • 1 Hide
    snotling , January 6, 2010 11:33 AM
    cjlBIOS update? Nope. The issue is drivers. You need to be able to load the appropriate drivers for the OS to see it.

    its not drivers either, its a matter of having a boot ROM on the addin card. something all products of this kind should have had from the start.
  • 1 Hide
    Anonymous , January 6, 2010 11:33 AM
    I'll tip my hat to Patrick and Achim for one of the first proper SSD reviews I've ever seem on Tom's Hardware. This article is head and shoulders above the others, both in terms of test procedure (finally accounting for used performance) and in the great variety of tests (I/O, random R/W, sustained R/W, and even some PCmark stuff).

    It's not perfect, but it's a big step in the right direction. I really hope to see more like this in the future.

    Cheers!

  • 0 Hide
    xtc28 , January 6, 2010 12:17 PM
    Another tip of the hat from me!!! Impressive article and deffinitely a piece of hardware I am looking into!
  • 0 Hide
    Pei-chen , January 6, 2010 12:35 PM
    Wow, that card is huge. Looks about 10.5" long at least.
  • 0 Hide
    zak_mckraken , January 6, 2010 12:53 PM
    "128MB SSDs with Indilinx controllers cost another $250 each."
    You probably meant 128GB.

    Seems like a great product but the price-point is still too high, even for enthusiasts. However, the future looks good for the average consumer!
  • 2 Hide
    jeffunit , January 6, 2010 12:59 PM
    I think they pulled their MTBF numbers out of their ear. I am unaware of any product with a fan with a MTBF of 1.5 million hours. A really good fan is good for 150 k hours. There are good ways to calculate MTBF. One way is to run many devices, like 100-1000 for a while to see the failure rate. Another is to calculate it based on each component. For example, the drive uses several SSD's. I have yet to see an SSD with a MTBF over 1.6 million hours. Putting 4 of them on a board, assuming everything else is perfect would yield a combined MTBF well under 1.5 million hours. A simplistic calculation would be 1.5/4 which is 375 k hours. So Super Talent clearly made up their number. If it were a real number, they would have a 5 year warranty. After all, 5 years is under 40 k hours.
  • 3 Hide
    jeffunit , January 6, 2010 1:05 PM
    How do you get 1.5 million hours MTBF with a fan?
    How do you get 1.5 million hours MTBF using 4 drives each having a MTBF of no more than 1.5 million hours.
    If you really have 1.5 million hours MTBF, why not have a 5 year warranty? After all, 5 years is under 44 k hours?

    It is clear that the MTBF number is pure vapor.
  • 0 Hide
    cadder , January 6, 2010 2:06 PM
    Why do you need a RAID mode with SSD's that provides redundancy? Are the drives unreliable like mechanical hard drives? It makes sense for striping in order to improve transfer rate.

    Also the article mentions battery backup. Why does a device with SSD's require battery backup?
  • 0 Hide
    vvhocare5 , January 6, 2010 4:57 PM
    One might need the battery backup to ensure all data was written to the drives - flushing the buffers and caches.

    This seems like overkill to me to use a x8 PCIe card which should be good for 4 GBytes of data transfer. A x4 card would still be overkill based on transfer capability. In most cases a Sata 6gbps might be enough for this.

    Im not sure I get the need for a PCIe slot based disk array? I can see the need if you have large DRAM caches that can stream data from multiple drives - but this solution doesnt go that fast. Add in the space constraints for an all in one style card and Im missing it. Server solutions seem to be faster - with external drives. Maybe its the reliability factor that is the differentiation with this product, but then you are talking SSD's which wont have a head crash and they back this up with a 1 year warranty??
  • 2 Hide
    GullLars , January 6, 2010 5:13 PM
    I think Supertalent made a huge mistake by not optimizing for IOPS. The Areca controller can easily handle 60-70K IOPS from an integrated RAID-0, wich is about what 4x Indilinx Barefoot drives can deliver for read.

    A guy on a norwegian forum i frequent (Nizzen, also on XS) has benched 5x Vertex in RAID-0 on his Areca 1680ix with 4GB RAM to the following 4KB random IOPS numbers with IOmeter at QD=32: Read 148,860 , Write 152,851. The price for this setup is just in excess of $2500, and with W7 on this array and his 24/7 clock settings he set a record for PCmark on ORB last spring, and still holds a top10 spot with the same setup.

    If you want to beat the living crap out if a RAID-drive, buy a LSI 9211-8i and run 8x Intel x25-V (40GB) in a dynamic RAID-0. This should produce more than 200K (240K with perfect RAID scaling) random read 4KB IOPS (QD>=32), 1500MB/s sequential read, 70K+ random write 4KB IOPS, and 320MB/s sequential write. This setup should cost no more than $250 + 8x $130 = $1290 (newegg prices 6 jan '10).
    To at all be able to utilize such storage performance you'd need a 4Ghz quadcore...
  • -2 Hide
    hakesterman , January 6, 2010 10:59 PM
    Until they can sell 512 Gig for $ 199.00 and 1 TB for $250.00 it's a waist of time period.............

  • -2 Hide
    hakesterman , January 6, 2010 11:07 PM
    CJI you are correct it needs a boot Rom and i was also right the Bios needs to know that the device cabability is there as well.


Display more comments