Seagate on NCQ

Our first look at an NCQ enabled drive was over a year ago with Maxtor's MaXLine III 250GB (NCQ) unit. Though not the highest in capacity at the time, the MaXLine III brought with it not only the NCQ feature, but also the largest buffer that we have seen in a desktop drive to date - 16MB. It outperformed the Hitachi Deskstar 7K400 in our IPEAK business and content creation Winstone 2004, SYSMark 2004 synthetic benchmarks, as well as in real world application performance and came second only to Western Digital's 10,000RPM 74GB Raptor. The unit did not perform as well as we thought that it would in our multitasking portion of our benchmarks, but its overall performance was exceptional.

Since then, we have only seen one other manufacturer implementing the NCQ feature in its drives and that is Seagate Technologies. One reason for this is the lack of support from motherboard chipsets for the feature. At this time, only Intel's and NVIDIA's newest motherboard chipsets support Native Command queuing, which limits the combinations of hardware that can be used. More specifically, boards like Intel's 910, 915, and 925 chipsets with ICH6 (I/O Controller Hub 6), and NVIDIA's newest nForce4 based motherboards, which have the Advanced Host Controller Interface (AHCI), are the only boards that will allow users to enable NCQ for drives with the feature.

How does NCQ work?

Native Command Queuing is a method of reading and writing to the disk, which takes into account the physical location of the list of requested data on the platters themselves. We described how NCQ works in our Maxtor MaXLine III review last year with our analogy containing the errand run between the grocery store, drug store, and mall, but we will again explain the process that an NCQ enabled drive goes through as well as the motive behind implementing such a technology in a desktop drive.

We all know how frustrating it can be to run multiple applications at once only to find that our multitask-capable OS slows to a crawl, especially those that require a large amount of reads and writes to the hard disk drive. Things slow down because as each application makes a request to access the drive, those requests are put into a queue and the drive will get to them on a "first come, first serve" basis. This becomes a problem when we have data scattered all over each platter, and the only way to retrieve this data is to wait for it to queue up; and considering that the list of requests is in a random order, this can take much longer than just a few milliseconds (very long when it comes to computing)!

Now, take your applications running on your OS and throw an NCQ enabled drive in the mix. The function of NCQ is all in the name. Native Command Queuing takes the randomized list of requests in the drive's queue and organizes each request based on the location of the requested data on the disk. For example, we have a 30MB Photoshop CS image on our hard drive, which is broken up into five segments on platter 1. The file is spread out on 5 different tracks and to make things more complicated, they are not tracks 1 through 5, but rather 1, 3, 5, 6, and 8, and each piece of the file is not in order on those tracks. When the application requests the PSD file, it sends to the queue of the hard disk the segments that it requires in order from A through E. To get from file segment A to segment B, the read head must skip from track 1 to track 5 to retrieve the data in that order. The read head must then move back to track 3 to pick up segment C of our 30MB PSD file and so on, and this skipping back and forth between tracks takes more time than it should. A more efficient way to gather the requested data would be to read from track 1, then 3, then 5, 6, and 8 last no matter what file fragment is picked up. The file can then be put together in the drive's cache for delivery. This is what NCQ does. As a file is requested, the NCQ feature organizes the five segments of the 30MB PSD file by their locations on the hard disk drive. After the drive reads the data, it is sent off to the OS. Much more efficient, right?

Of course, an NCQ drive can still operate with the NCQ feature turned off, but it will perform just as any other drive with randomized request lists. As we mentioned earlier, not all motherboard/chipset combinations offer support for NCQ, so those of you with older motherboards may need to pick up a newer Intel ICH6 or nForce4 board with AHCI support along with a new NCQ hard drive to take advantage of this feature. As we run our benchmarks, we will keep tabs on how NCQ helps both the Seagate 7200.8 and Maxtor's DiamondMax 10.

Index The RPM Factor
POST A COMMENT

44 Comments

View All Comments

  • zforgetaboutit - Thursday, May 26, 2005 - link

    The review has a table showing the drives' spec sheets. Among the stats are "average seek times (AST)". But I don't see average seek times benchmarked, as such.

    So, on the one hand, the Seagate's spec shows an AST of 8.x seconds, but other reviews have shown it to be 11+ seconds.

    I propose that if the review goes as far as to publish the purported AST, then it has an obligation to test it as well, with a discrete benchmark, such as HDTach or some other explicit AST benchmark.

    Otherwise companies will start to claim 2 ms AST, and Anandtech won't be able to refute it, if it's a blatantly bogus claim.

    Reply
  • OrSin - Tuesday, April 26, 2005 - link

    I see you explain how you take your sound measurements, but you really need to do it differently. The raptors are load as hell and seagate are queit and accourding to measure they are equal? Serious I undestand your reasoning for it, but it's just flawed. If a test seems right but produces obviously (and I mean obviously) wrong results then you need a new method.

    I had to send a raptor back it was so loud. I had to look at my computer (SFF) evertime i booted up to make sure it was going to rock off the table. Now my computer was a not actually moving but it sounding like it vibrating enough to move.
    Reply
  • Zak - Monday, April 25, 2005 - link

    Your articles are often difficult to read due to your use of some weird convoluted sentence structure. Why can't you guys use simpler, more accessible language??? Exampple:

    "RPM, or revolutions per minute, is the measure of instances that the motor of the hard drive can rotate the platters by a full 360 degrees."

    How about:

    "RPM, or revolutions per minute, the speed of platter rotation: how many times the platters rotate every minute." or something like that.

    Zak
    Reply
  • JPSJPS - Monday, April 25, 2005 - link

    Purav Sanghani - Poster 32 and especially poster 33 pointed out an obvious mistake that only a complete newbie would make. This makes all of your data questionable!!! Have you considered having someone with a little technical knowledge review your stuff before you publish it? Reply
  • PuravSanghani - Friday, April 22, 2005 - link

    TrogdorJW: The recordings as well as decibel readings were taken 1" away from the side of each drive. Obviously the sound emitted from the drives would not be as loud when inside a sealed case, but to get an accurate reading of the sound emissions from each drive and comparing them to each other requires that we take readings close to each unit.

    smn198: You are right, the frequency of the sound produced by each drive does make a world of difference. In the past when looking at case fans, we observed that larger 120mm fans are quieter than smaller 80mm fans because they produce a lower frequency which is less noticeable to humans. This is definitely the case with anything that produces any sound, including hard drives.
    Reply
  • ohnnyj - Friday, April 22, 2005 - link

    I want to know how you get Photoshop to open in under seven seconds on a Raptor. My RAID0 array opens in about 14-15. It opens faster if you open, close, then reopen again so I wonder if this is how the test was performed. Reply
  • Phantronius - Friday, April 22, 2005 - link

    Just bought 2 160gig Barracuda's to replace my noisey as hell Western Digitals. I freaking love them, soooooo much quieter.

    Reply
  • TrogdorJW - Thursday, April 21, 2005 - link

    A few things to note on noise (from my perspective):

    1) The hard drive noise levels were probably taken very close to the drives in order to capture them. Just FYI. 52 to 54 dBA is rather loud. Purav, what was the distance of your SPL meter from the drives?

    2) Seek noise can be very noticeable. My own experience reflects what's in the charts, with the Samsung being the quietest. Seagate and Hitachi are moderately loud, and the Maxtor and Western Digital Raptor are the loudest. (Raptor seek noise sounds louder to me than what's in the recordings.) I don't know about the decibel ratings, but it seems like if you started the charts at 50 that it would reflect more what I hear. (i.e. Samsung would be 1.2 to 2.4 and Hitachi would be 1.6 to 4.4)

    3) Bearing noise is generally either near-silent (Samsung, Seagate, and just about any other FDB implementation, including the Raptors) or else it's noticeable. The WDxxxxBB/JB models are notorious for having a lot of bearing noise, as are older Maxtor drives. I've got four WD's at my place, and I hate them all! :-p (They're a bit faster in a lot of tests, but they're too noisy.)

    4) Ignoring the echo in the recordings (MP3 compression can cause some funky artifacts), listen to the Maxtor 10 - Ouch! That thing is killer loud.
    Reply
  • JonB - Thursday, April 21, 2005 - link

    Since my motherboard SATA controllers don't support NCQ, I have three choices. Leave it disabled (which doesn't seem so bad in some respects), get a new motherboard, or add a Promise or Adaptec PCI controller card.

    Anybody got experience with an add-in Promise RAID with NCQ support???
    Reply
  • smn198 - Thursday, April 21, 2005 - link

    #28 - "Maybe we're seeing no boost with NCQ because of poor implementation, who knows. Testing with just one platform will not reveal such issues."

    I seem to remember Anand saying the opposite
    http://www.anandtech.com/cpuchipsets/showdoc.aspx?...
    "What's truly impressive, however, is the reduction in average response time - up to a 90ms decrease in response time, thanks to NVIDIA's superior NCQ implementation. "

    However, Anand did mention that NVIDIA took the decision not to 'turn on' NCQ until the queue depth had exceeded a certain amount. (Cannot remember which article that was in.) It may be that in some of these tests, this queue depth was not exceeded.

    #30 - "How is the 7200.7 120Gb drive louder then a Raptor? My 7200.7 120Gb drive is near SILENT, no where loud as a Raptor. I think your measuring device is off forthe Acoustics test."

    This may be due to the fact that the noise the Raptor emits is at a different, more audible frequency.
    Reply

Log in

Don't have an account? Sign up now