Performance Over Time and TRIM

Usually we use HD Tach to test TRIM performance but since SandForce drives do real-time compression, using HD Tach and compressible data would not give us a good idea of worst case performance. Instead, I filled the drive with incompressible data and proceeded to hammer it with incompressible 4KB random writes (QD32) for 60 minutes. Then I ran AS-SSD which uses incompressible data as well to see how the SX900 performs in worst case state. Finally I TRIMed the drive and ran AS-SSD again to test whether TRIM recovers the performance.

ADATA XPG SX900 - Resiliency - AS SSD Sequential Write Speed - 6Gbps
  Clean After Torture After TRIM
ADATA XPG SX900 128GB 175.4MB/s 35.4MB/s 146.8MB/s
ADATA S511 120GB 160.8MB/s 96.4MB/s 98.8MB/s

There is quite a big difference between the two ADATA drives when it comes to TRIM and post-torture performance. The S511 performs much better after torture but TRIM is not very effective. The SX900, on the other hand, is the complete opposite. Its performance is poor after torture but a single TRIM pass will almost restore the performance back to clean state, as after TRIM the performance is back to 84% of clean performance. The S511 is definitely an odd case because for example OCZ Vertex 3 behaves similarly with the SX900. However, the important part is that TRIM is able to restore the performance, which is what matters. Firmware is the apparent culprit once again, and it's unfortunate that the S511 hasn't been updated in quite some time.

Again, what I would like to emphasize that our torture scenario is an extreme case. In the real world, it's impossible to create a such scenario if you are using the SSD as a boot drive. That's because you will at least have Windows or some other easily compressible OS on the drive. Even if the drive was used as a storage drive, it's still fairly unlikely that you would fill it with incompressible data only, and what's more it is highly unlikely that a majority of your writes would be random in nature.

AnandTech Storage Bench 2011 - Light Workload Power Consumption
POST A COMMENT

58 Comments

View All Comments

  • XZerg - Friday, June 08, 2012 - link

    But at least include Mushkin in the benches, after all that's the cheapest SSD out there. This would give a better picture of $/performance. Reply
  • Kristian Vättö - Friday, June 08, 2012 - link

    We don't have any Mushkin drives in our labs. I have asked Mushkin for a review sample but I have yet to receive a reply. Reply
  • CeriseCogburn - Monday, June 11, 2012 - link

    I hate to say it but "good !" - that's what keeps their asynchonus prices so low. :-) Reply
  • Belard - Friday, June 08, 2012 - link

    Considering the history and being a favorite, I think the Intel X25-M G2 120GB should be included in the benchmarks. So many of us can track the improvements and see if its a good time to upgrade or not.

    The G2 still does pretty good for its slow-interface, since its not a SF drive.
    Reply
  • Kristian Vättö - Friday, June 08, 2012 - link

    Bench also works :-)

    http://www.anandtech.com/bench/Product/605?vs=126
    Reply
  • Belard - Monday, June 11, 2012 - link

    Thanks... Its still easy to forget that function is on the website. Reply
  • Assimilator87 - Friday, June 08, 2012 - link

    There's no point in including more than two, MAYBE three SandForce drives. One with synchronous NAND, one with asynchronous, and the Intel cuz it has a different firmware. They're all the same hardware with a different sticker. Reply
  • iceman98343 - Friday, June 08, 2012 - link

    when are you going to rereview vertex 4? new fw was released. Reply
  • UltraTech79 - Friday, June 08, 2012 - link

    Where are you guys getting these numbers? One place? The suggested retail price? The 256GB M4 is going for $0.70 per GB at amazon. Reply
  • Kristian Vättö - Friday, June 08, 2012 - link

    We used DynamiteDeals to hunt for best deals. However, the table is a bit outdated (the prices are from May 22nd). I know it should be more up-to-date, but the problem is that DynamiteDeals does not work for me as it's tied to one's IP address. Since I'm located in Finland, it only finds Finnish stores (well, there is only one it finds...). Jarred made the pricing table but in the end, Anand was the one who posted this article (it's been ready for ages, just needed Anand's final look at it). Reply

Log in

Don't have an account? Sign up now