The Performance Equation

USB flash drives are really quite simple devices. There are three parts to each USB drive that can actually impact performance: the flash controller, the flash memory itself, and the PCB on which the former two items are mounted. We're not talking about extremely high speed transfer rates here, so layout isn't as difficult or critical as it is with things such as motherboards, but it is a factor in influencing performance, albeit the least visible one.


Pictured here is the PCB of a USB flash drive. The left-most chip is a Samsung NAND flash device, and the other chip is a USBest flash controller. The connector on the right is a standard USB connector.

The problem with USB flash drives, however, is that the controller and the flash memory are rarely constant within the manufacturing of a single drive. In order to keep costs down and maintain a steady supply, most manufacturers obtain their controllers and memory from multiple sources, so buying one drive doesn't necessarily mean that you'll get the same combination of controller and memory that we test here.

Luckily, for the most part, this only applies to the lower end and mid-range flash drives. Even then, manufacturers guarantee some minimum level of performance, generally described somewhere on the packaging itself. But because of this variation in chip suppliers and/or types, performance will vary more than what we're used to between models of the exact same product. The higher end performance-optimized drives don't usually have this problem because the manufacturers put performance on a higher pedestal, thus they will always choose the same high performing chips to be used in these offerings. The manufacturing quantities are also much lower for the high-end, high-performance units, so having multiple sources for the controller and flash memories isn't as important.

Not only do manufacturers use different controllers/flash memories for the exact same product, but they do so at various storage sizes of that product line. For example, a 512MB USB flash drive from Company X may use a completely different controller than a 2GB drive of the same product family. Obviously, the flash memory itself is going to be different across different storage sizes, but do keep in mind that unlike hard drives, greater storage density does not necessarily mean greater performance; in other words, although different sized drives may perform differently, it isn't because of the size of the drive, but rather the actual components used on that particular drive.

There are a number of cases where the 512MB versions of a particular USB flash drive are faster than their 1GB and 2GB brethren, and other cases where the opposite is true. So, while we did our best to compare performance of equal sized drives in this roundup, note that the performance that you'll see here doesn't always map to larger drives. It is worth mentioning that although different sized drives do perform differently, the differences in performance are usually not enough to change dramatically the performance standings that you'll see in our charts.

Index Dual Channel Flash Drives
POST A COMMENT

37 Comments

View All Comments

  • BJL - Monday, October 16, 2006 - link

    Do the read and write speeds change for the 1gb and above models? Would I get the same performance, or should I stick with the 512mb? Reply
  • NeoZGeo - Monday, October 17, 2005 - link

    what kinf of benchmark are you guys using? How come some of those drive write speed is sooooo low!? Like Trenscent, OCZ Rally, i've seen some reviews out there which says other wise. Reply
  • NeoZGeo - Monday, October 17, 2005 - link

    here's the review by tom's

    if you look at the trenscend jet flash, it actually has the highest write speed average out at 23.3 mb/s vs anandtech's 8.7 mb/s !? what the hell?
    Reply
  • NeoZGeo - Monday, October 17, 2005 - link

    haha oops, forgot about the link :D

    http://www.tomshardware.com/storage/20050520/usb_f...">linky
    Reply
  • quanta - Wednesday, December 07, 2005 - link

    It looked like Tom's test is testing the write speed between USB host and flash drive's controller's memory buffer, instead of actual write speed, which can only be verify by doing a read after writing is completed. There are also reports that http://www.auphanonline.com/articles/view.php?arti...">cluster size may affect the write access behaviour. BTW, Tom's http://www.tomshardware.com/2005/08/10/two_fast_an...">later review has simultaneous reads and writes benchmark, which slows Memina Rocket to a halt. Reply
  • quanta - Wednesday, December 07, 2005 - link

    In addition, even when using buffered write in SiSoft Sandra, it is extremely unlikely that Transcend Jetflash 110 can write anywhere near 23.3MB/s. http://www.oc.com.tw/article/0510/readparticle.asp...">This benchmark shows that when doing random write with Kingston DataTraveler ELITE, write speed dropped more than a half compared to sequential write. Reply
  • TrueWisdom - Wednesday, October 12, 2005 - link

    I'm the in-house support for a university building and I've had absolutely horrible luck with Lexar drives. Lexar often fails to detect entirely on somewhat older machines (the Latitude C840, for example) and has also displayed a relatively high failure rate for me. I don't have any positive or negative impressions of Kingston drives, so I can't say anything there, but I will say that I've had by far the best luck with Sandisk drives. I've never had one fail on me, and I've seen them go through wash cycles, get run over by a car, and get left out in the rain. Compatibility has been universal as well. They may not be the fastest drives but they really are the only ones I've ever trusted. Reply
  • pendrivethis - Friday, October 07, 2005 - link

    i work for a flash memory controller maker and in all honesty the most meaningful performance test is random write. and no one really advertise that since sequential read speeds seem much more appealing and marketable. i can get a dual-channel & interleaving enabled usb 2.0 pen drive with micron or samsung nand-type flash to go up to 34mB/s in sequential read, but the engineer who designed this still tells me that he'd rather use and-type flash from renesas (formerly hitachi) since and flash has a better random write than nand flash.

    and knowing what i know, if you use your pen drive very often, and i suspect some of you may be in that boat, i'd check out some of the burn-in testing results especially since companies are not entirely using only samsung nand flash. certain new flash whether nand-type or ag-and-type and even some high-density samsung flash seem to be displaying a need for extra care in ecc. data-verify errors are fatal, especially if it's the only copy you got...

    anand, perhaps a little visit to some of the design houses for these controllers the next time when you're in taiwan is in order. computex is only 9 months away.
    Reply
  • sprockkets - Thursday, October 06, 2005 - link

    Although not reviewed here, I got the A Data key from newegg.com simply because they say it works with Linux on the package. I know any key will, but they are the only ones to have the guts to say it. Thanks for admitting Linux exists. Lifetime warranty too. Reply
  • jgh - Wednesday, October 05, 2005 - link

    here is a link for another link, to an app that can make many (but probably not all) usb drives bootable and a couple of other hints/tips.

    http://www.oreillynet.com/pub/wlg/5735">link

    O.T. - for some reason i get a message that says i do not have permission to access this forum when i tried to create a new login with my e-mail address. did i get banned or something? i have only posted once (it was about the gta:lcs website). i also cannot log in with the origianl user name and password.

    p.s. - it is o.k. to post links like this right?
    Reply

Log in

Don't have an account? Sign up now