Performance Over Time & TRIM

SandForce has always exhibited strange behavior when it came to TRIM. Even Intel's custom firmware in the SSD 520 wasn't able to fix SandForce's TRIM problem. The issue happens when the SSD is completely filled with incompressible data (both user LBAs and spare area). Any performance degradation after that point won't be restored with a TRIM pass and instead will require a secure erase to return to new. I didn't expect the 335 to fix this but I still tortured the SSD 335 for 60 minutes, ran AS-SSD, TRIMed and reran AS-SSD:

Intel SSD 335 - Resiliency - AS SSD Sequential Write Speed - 6Gbps
  Clean After Torture (60min) After TRIM
Intel SSD 335 240GB 317.7MB/s 174.2MB/s 176.9MB/s

And the issues persists. This is really a big problem with SandForce drives if you're going to store lots of incompressible data (such as MP3s, H.264 videos and other highly compressed formats) because sequential speeds may suffer even more in the long run. As an OS drive the SSD 335 will do just fine since it won't be full of incompressible data, but I would recommend buying something non-SandForce if the main use will be storage of incompressible data.

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

69 Comments

View All Comments

  • meloz - Tuesday, October 30, 2012 - link

    >After Anandtech's 840 Pro review sample died, I'm not super inspired with confidence about the plain 840.

    Oops, I had almost forgotten about that, thanks for the reminder.

    Damn, all SSD manufacturers suck in their own ways.
    Reply
  • jwilliams4200 - Thursday, November 1, 2012 - link

    I'm disappointed that there hasn't been any more information on that 840 Pro that died.

    Anand should really post some more details. Like what it was doing just before it died, the symptoms of how it failed, whether the SMART parameters could still be read, etc.

    Also, Anand should be hounding Samsung to get back to him about it, if they haven't already. The 840 Pro is apparently shipping on Nov 6. If Samsung has not been able to diagnose the problem and report back by then, it looks bad for Samsung.
    Reply
  • Kristian Vättö - Thursday, November 1, 2012 - link

    Anand was filling the drive with sequential data (preconditioning it for our enterprise tests) and it just died in the middle of the run. After that it was no longer recognized in BIOS, not even when connected using USB to SATA adapter.

    As far as I know, Samsung has not gotten back to us about it yet but let me ask Anand and see if he knows more.
    Reply
  • jwilliams4200 - Thursday, November 1, 2012 - link

    Thank you for the additional information.

    I am highly interested in what Samsung has to say about the failure. It seems to me that anandtech should be able to put some pressure on Samsung to give them a thorough failure analysis in a timely manner, or else anandtech will report that Samsung was unable to explain the failure and that looks bad for Samsung.
    Reply
  • mmonnin03 - Tuesday, October 30, 2012 - link

    The flash on SSDs arent going to get more reliable. ECC basically scales exponentially as the process dimensions keep shrinking. As the lines get closer and closer, the number of electrons holding the charge becomes harder and harder to measure. Each cell is 2 bits, so 4 different amounts of electrons need to be measured. Errors occur more frequently and get fixed. And that's not going to make the NAND any faster by going smaller. SSD speed/reliability improvements will/have come at the controller level. If you truly want a reliable SSD, go 34nm SLC. Its still being produced. Reply
  • jeffrey - Monday, October 29, 2012 - link

    Between the Samsung 840 Pro dying during testing and now the endurance issue testing Intel's 335, I believe OCZ should get some more credit for their Vertex 4.

    Everyone wants to mention Intel, Micron/Crucial, and Samsung for SSDs, but the Vertex 4 deserves to be there too.
    Reply
  • josephjpeters - Monday, October 29, 2012 - link

    Endurance will play a larger role in differentiating future SSD's as the industry continues to move to smaller NAND geometries.

    I'm interested in seeing OCZ's Vector which will use 20nm MLC NAND. It'll be a big test for OCZ to see how their endurance technology stacks up against the competition (or lack thereof in the consumer space).
    Reply
  • Samus - Monday, October 29, 2012 - link

    I've had my Intel SSD320 160GB for about two years, use it daily, and still have only written 10.6TB to it and the MWI is still 100%.

    Either way you look at it, this drive will last the average user a decade easily, even with less than 1000 P/E
    Reply
  • IntelUser2000 - Tuesday, October 30, 2012 - link

    This is false: "Based on the data I gathered, the MWI would hit 0 after around 250TB of NAND writes, which translates to less than 1,000 P/E cycles."

    There is a forum that exclusively tests SSD endurance and Intel drives last far past the MWI of 0. In fact, after it reached zero, it started counting up. I remember the original X25-M lasting until the second MWI is significantly greater than 25(it could be 50 I don't remember).

    They thought that after the Media Wearout Indicator reached 0, the drive would die. In fact, none of the drives did. NONE.

    Even 240TB is hell of a lot. My X25-M has 7.6TB written to it and I had it since the year the drive was announced. At this rate, I'll be 30 years older by the time it reaches that point. So its a needless worry about nothing.

    Contrary to Platter HDDs, which die off slowly and more and more data gets corrupted and gets slower and slower until you notice that the drive is dying. Less than 5 years for lot of people around me too.
    Reply
  • Kristian Vättö - Tuesday, October 30, 2012 - link

    I did not say the drive will die after the MWI hits 1. In fact, I said the opposite:

    "Even after it has hit 1, it's likely the drive can still withstand additional write/erase cycles thanks to MLC NAND typically behaving better than the worst-case estimates."

    The problem here isn't that 1,000 P/E cycles isn't enough for a consumer, but the fact that there seems to be a huge difference in endurance between 20nm MLC and 25nm MLC if our data is correct. Intel claimed that there is no difference, both are 3,000 P/E cycles, but our data contradicts with theirs. Given that the SSD 335 doesn't bring any immediate price cuts, you are getting a worse product for the same money compared to the SSD 330.

    It's of course possible that there is a simple firmware bug which reports wrong MWI or NAND writes, but at least so far Intel has not said anything to suggest that.
    Reply

Log in

Don't have an account? Sign up now