back to article Why is USB 3 so slow?

USB 3.0's SuperSpeed rating is a dismal joke if magazine tests are anything to go by, with transfers at a laggardly 127MB/sec at best, only three to four times faster than good USB 2 products. So why is USB 3.0 so slow? In theory, USB 2.0 runs at a maximum of 480Mbit/s, whereas USB 3.0 runs up to 4.8Gbit/s: ten times faster. …

COMMENTS

This topic is closed for new posts.
  1. Inachu
    Heart

    I use all my USB ports!

    I have 6 usb ports on an old pc that is dog slow!

    I plug each usb up with a usb 2.0 thumb drive.

    I put a swap file onto each one with min/max size of 512 megabytes.

    Reboot.

    now the 500Mhz pc acts just as fast as a 1.4Ghz pc :-)

  2. Steven Raith
    WTF?

    It's the spindle, stupid.

    Rotational drives - especially consumer level - rarely get beyond 130MB/sec as a peak transfer rate. You then have to factor in the SATA-->USB bridge chip, and also the stated problems of USB having to potentially schedule for up to 127 devices, etc.

    Now, if someone could pull apart one of these devices and throw an Intel X25, or other performance SSD [most SATA-->USB bridges will work with any SATA device] then you might get an idea of where the *real* bottleneck starts coming in.

    As for multi-disk USB drives - I'd suspect most of these will end up having a Gigabit NIC on them too, seeing as the cost of prepping RAID-capable SATA-->USB bridges will soak up most of the overhead - once you get to that sort of price point, you might as well start looking at adding a NIC, letting an ARM SOC with a slim linux distro do all the RAID/LAN/WebUI work.

    It concerns me that people don't seem to realise that the limitation with a multiple gigabit link on USB hard drives is the disk though - especially as most HDDs can't even saturate SATA II at the moment, never mind SATA III.

    Steven R

  3. jason 7
    Thumb Up

    127 Mbps?

    Hell after being stuck with around 18Mbps for USB2 for so long I'd have settled for 36Mbps to be honest.

    127 Mbps will be a nice boost, I can live with that.

    What does annoy me (and I never take stock in it) are these ludicrous theoretical maximums the labs come up with. I usually dived by 5 and that sets my maximum expectations.

  4. Anonymous Coward
    Anonymous Coward

    Excuses, excuses

    No, I don't buy it. Either you have a good design and the speeds become a breeze, or you're stacking botch upon botch and you lag more and more behind your maximum "theoretical" speed. It's not like connecting things hasn't been done before. Ethernet springs to mind, even though there is very little left except the embarrasingly low frame size from the original. If you want performance you'll have better chances with eSATA, or dare I say it, for it is true, firewire. To me, USB 3.0 is marketing numbers like the pentium 4 lagged considerably behind the pentium III on a clocks per MHz basis, everything sacrificed for sheer MHz, to improve the numbers marketeering could throw at the masses. It's in good company, though: 802.11n springs to mind.

  5. Gerard Krupa
    Dead Vulture

    Apples and oranges

    Why are you comparing observed maxima for USB 3 with theoretical maxima for USB 2? In fact why does this read more like an advert (or maybe an apology) for Buffalo than an article about USB speeds?

  6. Anonymous Coward
    WTF?

    ...+ a quad core required

    The thing thats forgotten is how shat USB is when it comes to computer resources (specifically CPU).

    It is one of the worst inventions for anything beyond keyboards/mice.

    It sucks the life out of your PC ... why would you use it for high value I/O ? Use a real I/O controller and you can continue using your PC in the meantime!!!!

  7. Russell Howe
    FAIL

    120Mbyte/s device fails to saturate 400Mbyte/s interconnect

    ... shocker!

    1. Anonymous Coward
      WTF?

      *ding*

      Old bottleneck no longer bottleneck.

      Hold the front page.

  8. edwardecl

    Agreed USB does batter your CPU

    Or at least USB 1 and 2 did, no idea about 3 but I will assume it's the same. That's why I always liked firewire, sure it was more expensive but it didn't completely batter the CPU and used a normal voltage (12v), wish USB used 12v :(.

    1. M Gale

      But it is a standard voltage.

      5v is a standard TTL line voltage.

      Just so you know.

  9. Stu
    Boffin

    PCI/PCI-E bridge limits too?

    This is just a thought, I haven't done the maths, but wouldn't some of the slowdowns come too from the controller connection to the rest of the system, i.e. PCI or PCI Express x 1?

    If you have a brand new motherboard come out with USB3 on board, that too will be connected to the rest of the system via a PCI bus and suffer from all sorts of bus contention issues, lets face it there's a lot of inherent traffic moving around your typical PC all the time enough to eat into the benchmarks.

    I agree though that the slowdowns are most likely from the limitations inherent in those olde spinny platter devices.

    How about everybody stop bitching about USB3 until those chunky MLC SSDs start appearing with dedicated USB3 front ends? (So many TLAs)

    Ithankyou.

    1. Danny 14

      not really

      The pci bus is no longer the backbone in most pcs. The southbridge tends to control legacy buses including PCI and USB. There is a dedicated interconnect running faster between the CPU (NB or otherwise) and the SB (or otherwise) so there shouldnt be saturation.

  10. Frank Bough
    WTF?

    Faster than eSATA?

    127MB/sec?? I have an eSATA RAID attached to my machine here which exceeds 300MB/sec on a single eSATA PM link. I don't think Buffalo know what they're talking about.

    USB3 sound like a load of garbage to me - all the overhead of Firewire without the advantages, and barely faster than FW800 at the moment either.

    1. Tim Parker
      FAIL

      Re : Faster than eSATA

      Yeah - i'm having difficulty trying to work out WTF was meant there as well...

      "Reid is pleased with this 130MB/sec figure, partly because it's faster than eSATA, which, like SATA II, runs at 3Gbit/sec"

      ..and what is this trying to infer - that something with "up to" 130MB/s bandwidth (no details on that either) couldn't possibly be beaten by something with a peak, theoretical bandwidth of 375MB/s ? (assuming 8-bit bytes) Reid must know something about the SATA protocol overhead that the rest of us don't...

      Sorry - to me this is yet another storage report that smacks of aping a companies PR machine with little or no interference..

      1. Nexox Enigma

        @Tim Parker

        """..and what is this trying to infer - that something with "up to" 130MB/s bandwidth (no details on that either) couldn't possibly be beaten by something with a peak, theoretical bandwidth of 375MB/s ? (assuming 8-bit bytes)"""

        Indeed, that caught my attention too. Most people use a 10 bit byte (10 bits on the SATA for 1 byte of file, thanks to error correction etc) to get 300MB/s theoretical over SATA, which is still far faster than 130MB/s.

        I have a single (cheapish MLC) SSD that easily gets 180mb/s over a single sataII port, using a standard on-board sata controller.

        For those who are curious, the pci-e 1x speed is about 250MB/s, so a usb3 controller would need a 2x connection.

  11. Richard 102
    Thumb Up

    AC @14:35 GMT

    You are absolutely right; USB's spec ties performance to the CPU and system, thus making it depend on a lot of things other than itself. I'll stick with Firewire, where applicable.

  12. Sarah Davis
    FAIL

    Hardly suprising

    aren't Buffalo the Amstrad of the external drive world. I got a Buffalo Giganet NAS which at best performs like any good 10/100 NAS.

    Buffalo specs are unrelaible. They probably just got a USB2 device and put a USB sticker on it.

  13. Anonymous Coward
    Linux

    Could it be the OS?

    The Windows driver model is known to be atrocious, and there are vendors offering replacements for USB 3.0.

    As far as I could see on both of the quoted articles, the tests were probably run under Windows.

    The Reg should try a comparison between Windows and Linux, given the same hardware - after all, Intel did the initial development of USB3 under Linux.

  14. Gary F
    FAIL

    Not an intelligent article

    What a stupid article, whinging that a SATA-II hard drive at 7200 rpm doesn't fully use the USB3 bandwidth. What's so shocking about that? If you put a Ford Fiesta on the Silverstone racetrack would you be shocked that it doesn't hit 200mph on the straights?

    There is no way of saturating USB3 with a current domestic hard drive. Even a pair of 15,000 rpm SAS drives in RAID0 couldn't fill that much bandwidth. But if you have several USB3 hard drives attached then you would benefit from USB3 if they're all copying files simultaneously. Remember, USB3 is to satisfy computing for the next 6 years, so we won't see affordable devices capable of saturating it for a couple more years.

  15. Anonymous Coward
    Anonymous Coward

    Good Article Idea

    nice try, lie back and think of England, and all that. But, really everyone wants the real stats and identification of where the bottleneck is, maybe a table would help.

    There is more to the USB 3.0 story as well, and it looks like another BetaMax, VHS war is in the brewing. More on the Sata line would be better as well.

    Interesting stuff, but a bit more effort would be appreciated.

  16. Anonymous Coward
    Anonymous Coward

    USB controller complicated? Why?

    USB was designed to be a replacement for the serial port. A simpler interface you'd be hard pressed to find on a PC so why is USB so complex? So it has to handle 127 connected devices - BFD. All the controller has to do is act as a data format converter between the data bus and serial device - what is plugged in is irrelevant , thats for the OS to worry about at a higher protocol level.

  17. Anonymous Coward
    Linux

    @ Could it be the OS?

    How is it every article has to get back to Linux? Every flipping article you read has some loser say something about Linux! An article about USB 3 and we have some simpleton wanting to compare USB3 speeds on Windows and Linux.... listen no-one cares you understand? No-one cares about how it might perform on Linux. I hate the very mention of the word Linux at this point. Actually, most USB3 devices won't even support Linux, so I would not worry too much about it. In fact I'd be willing to bet Windows would outperform Linux, simply cos the devices will actually work on Windows without you having to hack for thirty minutes in the console to try and make the device mount.

    Please get me an icon where people can show their contempt for that useless, 0.05% market share grabbing, dog that is Linux.

    1. James Hughes 1

      Why not bring up Linux?

      Surely the point of driver efficiency is a very valid point and a comparison between Linux and Windows using USB 3 would be a good idea?

      Now, you obviously don't like Linux - that's of course your prerogative, but the author of the comment was making a completely valid point, and the fact you jumped down his throat in a Linux bashing and wholly irrelevant comment, sort of make you the loser, don't ya think?

      (And most USB3 devices won't support Linux? That comment make no sense whatsoever. Since when have the devices had to support the OS?)

  18. Stan 2

    bloody stupid article

    I'm using raid to get about 250MB/S, most disks do about 125MB/S. Sorry for repeating what plenty of others have said already but the article just doesn't make how these tests where done clear. Either solid state storage or direct from memory reads/writes would have given the bus a chance to do its stuff and, I'm sure, would have given much better results despite the theoretical figures being pie in the sky.

    Also, any chance of a 'return to comments' link after going through the thumbs up/down rigmarole?

    cheers

  19. jon 72
    Alien

    Garde Ta Foy

    Our dark alien overlords will release the better version, next Christmas perhaps .. if we are good

  20. Goat Jam
    Thumb Down

    USB sucks

    I'm using it at this very moment for what is, I grant you, a rather unusual project.

    Scenario. I have a PC with a CF->IDE "drive" and no optical drive. Normally, I would simply boot and install ubuntu off a USB key but for some reason this mobo refuses to boot Ubuntu in said manner. It is apparantly a BIOS problem (according to some googling)

    I want to install ubuntu to an 8GB flash card in that server.

    Anyway, my workaround is to stick the CF card into a USB card reader, fire up vmware workstation, build a VM with "Use a physical drive" instead of a virt disk and point that at the CF card.

    The process ultimately works fine. I can install the OS to the card, remove it and then boot it in the IDE CF drive bay all without problem.

    However, during the installation process using vmware, the PC becomes unusable. It locks hard (including mouse pointer) for minutes at a time. The first time it happened I almost did a hard reset thinking it had hung completely. Eventually it came good in spits and stutters but I had to walk away for over an hour until it was finished. An install which normally takes about 25 mins took 4 times as long and the host PC was unusable during the process.

    I've done similar installs using this method in the past using non usb drives without this problem. The problem is the USB interface. It is too CPU dependant. This makes USB cheap to produce but does not make it good.

  21. Christopher 5

    USB Really is NOT up to the JOB

    I am amazed at this history of USB....

    All these Points make no sense, saturate a USB III BUS, its a not about a single attachment, the second you share a bus of any type, massive slowdown is seen..

    I had a Keyboard, a mouse, and 2 external disk drives, and a Memory card reader.

    Performance is scatty, some seconds I get 12 MB/ second, next 1 MB/sec...

    Moved to Firewire, Boy I will never look back.....

  22. Matthew 4
    FAIL

    completely pointless

    In the real world of cheap flash memory etc etc people don't even utilise USB 2.0's full potentional so I really don't see the point

  23. greeno99

    Use firewire

    Firewire 800 - much faster than USB 2 and it just works!

  24. Another Anonymous Coward 1
    Thumb Down

    Needs an SSD to evaluate USB 3.0 speed.

    Or, to at least come close to it.

    This article is misleading... from the sounds of it there's nothing wrong with usb 3.0 or the controller and it's clearly the disk that can't physically output more than 130 MB/s.

    Calling USB 3.0 slow because the disk it's connected to can't fill up the pipe is a bit like saying my 20 Mb internet is slow because I just visited a site where the server hosting it couldn't output more than 26 kb/s. Completely the wrong end of the stick.

  25. John 61
    IT Angle

    127 devices on a USB controller??

    Erm, who's barmy idea was this? Even on my old HP effort (a P200 job) wich only had 2 ports on it, I only managed to connect 3 devices via a powered hub and only then I had only low bandwidth stuff connected to it. It worked fine with a bit of creative use. Of course if you have a massive desk and an unlimited amount of money and time you could daisy chain stuff, but it would be so slow it would be unworkable. It's a bit like public transport: you can't get on the bus if it's full.

  26. Henry Wertz 1 Gold badge

    USB3

    @AC, Linux has better USB support than Windows. The HD enclosures I've used are consistently faster under Linux than Windows. It also has support for a wider range of hardware (you won't find any one version of windows that supports the old Windows 98-era web cams *and* modern stuff, while Linux does.) Saying Linux won't support USB3 stuff is daft -- one of the whole points of USB3 is, other than supporting the actual USB3 controller, that stuff will be basically USB2 compatible but faster. I.e. neither Linux nor Windows actually require any other new drivers to support most USB3 stuff. Intel developed the USB3 support for Linux initially.

    That said, I think the hard drive in the enclosure hit it's speed limt, and Linux would get exactly the same speed. But come on, it's daft of you to not at least want to know; if Linux were faster, for instance, then even if you only use Windows you know you could get this hardware and possibly get a nice speed boost down the line (via driver improvement or whatever.)

    My record for USB is keyboard, mouse, printer, scanner, and 10 floppy drives. With 10 floppies. Which I then raided and put a bit of swap onto hahaha. They were those drives that sound a bit like can openers when they do a full stroke, the racket was hilarious. Length limit? We hooked about 90 feet worth of 6-foot extensions at my work and a mouse still worked (mice and keyboards usually run at like 1.5mbps instead of 12 or 480). A webcam did not work, it would only run about 30 feet.

  27. StorageGuy

    Check Your Math

    "...this 130MB/sec figure, partly because it's faster than eSATA, which, like SATA II, runs at 3Gbit/sec..."

    3Gbit/sec = 300MB/sec (approx.), so 130MB/sec is not faster than 3Gb eSATA

    And stay tuned for 6Gb/sec eSATA...

  28. Ardkin

    Too Much Bandwidth?

    @People comparing Linux/Windows

    Ok, I really don't get a stiffy at windows or Linux. All I know is we're currently developing a high bandwidth (>200Mbytes/sec) USB3.0 Isochronous device. Windows 7 says "I'm sorry, that's too much data for me to cope with" and Linux says "bring it on baby" when I plug it in.

    Problem is our market share is crippled supporting Linux only.

  29. Herby

    So slow? Because it isn't Firewire!

    Instead of USB-3, Intel should have promoted Firewire as the better alternative. I really don't need a 100Mbit.sec data rate for my keyboard or mouse.

    The better alternative was Apple's ADB bus. IT was designed for the small number of input devices (keyboards and mice) that were in use. USB is just another mish-mash that started out reasonable, but got more and more involved. Look, it even gave a blue screen for Bill Gates in a televised demonstration, so it didn't do its job when it was first implemented.

    Enough said.

    Of course, I could go on about the IDE interface vs. SCSI, but that is another kettle of fish (*SIGH*).

This topic is closed for new posts.