Crucial/Micron RealSSD C300 - The Closest Competitor

While OCZ rushes to be the first to ship these superfast SSDs, Crucial and Micron will soon be shipping their RealSSD C300s. Based on a Marvell controller these drives (Crucial for the channel, Micron for OEMs) are far more traditional in their architecture.

Instead, the innovation comes from the use of ONFi 2.0 MLC NAND flash and a 6Gbps SATA interface. The combination of the two results in some extremely high sequential speeds. A seemingly well architected firmware (and a boatload of DRAM) work together to deliver good random access performance as well.

In testing the C300 it performed very much like a faster X25-M, there was one anomaly that bothered me: maximum write latency.

Like Intel’s X25-M, whenever the C300 goes to write data it also does a bit of cleaning/reorganization of its internal data. The more cleaning the drive has to do, the longer this write process will take. Micron did its best to minimize this overhead but eventually you’ll have to pay the piper. Below you’ll see the average IOPS, average MB/s, average and max write latencies for the C300, X25-M G2 and Vertex LE during my 4KB random write test:

4KB Random Write Performance Average IOPS Average MB/s Average Latency Max Latency
Crucial RealSSD C300 36159 IOPS 141.3 MB/s 0.0827 ms 1277.9 ms
Intel X25-M G2 11773 IOPS 46.0 MB/s 0.255 ms 282.9 ms
OCZ Vertex LE 41523 IOPS 162.2 MB/s 0.072 ms 109 ms

 

While both Crucial and OCZ/SandForce offer incredible average write latencies, Crucial’s max latency is over a second! I haven’t actually seen max write latencies this bad since the JMicron days. But if you look at the average write latency, you’ll see that this max latency scenario basically never happens. I only worry about what happens when it does.

Crucial also warned me that despite the controller’s desire to keep performance as high as possible, if I keep bombarding it with random writes and never let up it may reach a point where it can no longer restore performance to an acceptable level. This sounds a lot like what Intel encountered with the original X25-M bug, although it’s not something I was able to bring about in normal usage thus far. Given the early nature of many of these drives, it’s going to take a lot of consistent use to figure out all of their quirks.

Overall performance of the C300 is excellent. Just like the Vertex LE, it performed admirably in all of our tests. Paired with a 6Gbps controller there’s actual a noticeable improvement in real world performance, although it’s limited to those scenarios where you’re doing a lot of sequential reads from the drive.


6Gbps SATA controller on a PCIe x1 card

The drive’s performance does come at a price. The RealSSD C300 will be available later this month in 128GB and 256GB configurations, priced at $499 and $799 respectively.

OWC - Hot on OCZ’s Heels What’s Intel Doing?
POST A COMMENT

83 Comments

View All Comments

  • allessd - Saturday, February 20, 2010 - link

    maybe this site is of interest to you:

    "SSD Decoder Ring - an SSD comparison guide"
    www.pcper.com/article.php?aid=736
    Reply
  • Jupie - Saturday, February 20, 2010 - link

    Great article - always nice to hear about SSDs as I clearly see them as the future for a boot drive at least.
    What I'm curious about is the performance of drives of the same series with different capacities. I'm not willing to spend ~800 USD on a drive but rather ~400 USD. How much performance hit does a Real SSD C300 really take? Manufactors clearly seem to prefer to send their biggest drives with the highest performance to test but what about the performance of the reasonalbe priced ~100 GB versions?

    Would be great to read about that as well ;o)
    Otherwise keep up the good articles!
    Reply
  • MadMan007 - Saturday, February 20, 2010 - link

    Yeah this is an important point, it applies to regular HDs to a much lesser extent too. I would like to see tests on smaller capacity drives of a given series. I guess a lot depends on implentation, for example using the same number of flash chips with a given controller just of a lower capacity per chip versus using fewer flash chips and channels. Reply
  • Conscript - Friday, February 19, 2010 - link

    Not sure why I'd pay the premium for "limited" SF-1500 drives when I can just get the same thing here, for less...

    http://eshop.macsales.com/shop/internal_storage/Me...">http://eshop.macsales.com/shop/internal_storage/Me...

    Anand, any chance you think they'd give you one for review? Thinking about putting one of these in my new MBP (when they come out). Unless the new MBP has SATA 3, in which case I might look hard at the C300.
    Reply
  • iwodo - Friday, February 19, 2010 - link

    May be Anand can do an article on that. You mention previously that controller contribute VERY little to the total cost of SSD. NAND being the major part.

    If that is the case, then assuming we always need 8 chips for an SSD. SSD Prices wont ever dropped below $100? Because while capacity per unit will grow, it seems minimum unit prices dont fall that much.
    Reply
  • icrf - Saturday, February 20, 2010 - link

    Yeah, I'd love to see a reliable, slow 8-16 GB drive for $50. I've got a RAID server that can't seem to keep up boot drives and would love something like that. Reply
  • Bolas - Friday, February 19, 2010 - link

    So if I'm willing to spend up to $1200 for a boot drive, what's the best option? Crucial? Vertex LE? Intel? OCZ Z-Drive? Something else? Reply
  • czesiu - Friday, February 19, 2010 - link

    Any chances for a Kingston SNV425 review? Reply
  • mckirkus - Friday, February 19, 2010 - link

    I would like to see what pure RAM drive based storage would do so we can get a sense of how close these drives are getting to some sort of maximum.

    I'm not sure if you can do this with a RAMDisk you create with software or if and entire image would need to run in RAM (HDTach, etc., will not work on RAM drives for some reason. Just to be clear, I'm not referring to benchmarks of RAM using a SATA interface.

    Good stuff yet again Anand.
    Reply
  • mindless1 - Friday, February 19, 2010 - link

    There is no "sort of maximum", DRAM keeps getting faster.

    However, if you were to create and use a RAMDISK, however you ended up doing so, it would destroy any flash drive in terms of performance on every possible parameter. The question is not about performance at all, rather the implementation and volatile nature of using it... and of course that it's no small feat to end up with hundreds of GB worth of those chips for typical HDD replacement purposes (in a reasonably sized form factor).
    Reply

Log in

Don't have an account? Sign up now