USB 3 in 2008, 10 Times as Fast

psychicsword writes "Intel and others plan to release a new version of the ubiquitous Universal Serial Bus technology in the first half of 2008, a revamp the chipmaker said will make data transfer rates more than 10 times as fast by adding fiber-optic links alongside the traditional copper wires." "The current USB 2.0 version has a top data-transfer rate of 480 megabits per second, so a tenfold increase would be 4.8 gigabits per second." This should make USB hard drives easier and faster to use."

I'm more concerned with latency.

[jimstapleton]

It seems current hard drives test to 40-80M/s (dunno if it's bit or byte, we'll assume byte since it is worst case for my example)., averaging between 50 and 60M/s

480Mbit per second = 60MByte per second. That can handle the average case for a modern hard drive.

4.8GBit/second - 600MByte/s? To utilize that with a drive, you'd need a RAID external enclosure!

Re:Cable?

[YrWrstNtmr]

The last line in the article:
"It will be backward compatible, so current USB 2.0 devices will be able to plug into USB 3.0 ports."

Re:Great.

[jimstapleton]

1.5 seconds if you all of your components were fast enough. The drive won't be.

Re:Great.

[morgan_greywolf]

1.5 seconds if you all of your components were fast enough. The drive won't be.

Exactly. There's still the slowdown associated with the mechanical aspects of the hard drive -- spin rate (RPM), average seek time (ms), etc. On top of that, most hard drive controllers are limited by the technology they use. For instance, a SATA hard drive, even plugged into a USB 2 or 3 port, is limited to 150 MB/s -- but, that's burst speed, not sustained transfer rate.

Re:I'm more concerned with latency.

[Televiper2000]

My experience with USB is that it has lousy thru-put do to the way the drivers manage data on the host side. We did some heuristic timing tests on USB to serial devices and found that most of them actually degraded the thru-put (time send and receive a packet) of an RS232 connection. We found that devices with Silicon labs ICs could get the desired thru-put, but it seemed they would take up an entire USB channel to do so.

Re:Great.

[Anonymous Brave Guy]

On top of that, most hard drive controllers are limited by the technology they use. For instance, a SATA hard drive, even plugged into a USB 2 or 3 port, is limited to 150 MB/s -- but, that's burst speed, not sustained transfer rate.

Indeed. And realistically, it's going to be a pretty short burst: most hard drives today only have something like 8–16MB of cache that might be filled by a smart lookahead algorithm, so your best case with current hard drive technology is that you'll get perhaps 1/10 of a second of high-speed data transfer before hitting the physical barriers.

I'm not sure this is directly applicable to this discussion, though, because AFAIK all current USB drives use different storage technology anyway. It's going to be the limits of that technology that tell us whether USB3's theoretical speeds will actually be useful with storage hardware available in the same time frame.

Re:Screw bandwidth...

[jimicus]

The problem with that is that a PSU for a powered 7-port USB hub already has to supply 3.5 amps. Much larger than that and with current technology you're getting into "big expensive PSU" territory.

Re:Great.

[pipatron]

Uh, what? 0.9TB/(4.8Gb/s) = 25.6 minutes, not 1.5 seconds, even IF the platters would cope with the new speed.

Re:Great.

[Retric]

Almost but 4.8 gigabits/s = .6gigabytes/s .9TB = 900 gigabytes
900 gigabytes / (.6 gigabytes/s) = 1500s or 25min.

Re:Great.

[mikael]

You can get an USB to IDE adapter in order to convert a 2.5" IDE laptop drive into an external hard drive. It just clips over the ID adapter and plugs into a USB cable.

Re:Great.

[jonnythan]

Yeah, this one is correct. I divided by 480Gb/s instead of 4.8Gb/s because I glanced at the summary and saw "480."

It's not 15 seconds, it's 1500 seconds.

So much for being able to do basic math.

GP forgot that a byte is 8 bits.

Re:Bottleneck?

[Garabito]

As far as I know, the term 'Synchronous' refers to the use the same clock signal at both ends of the communication link to synchronize the data transfer; in that sense, USB wouldn't be synchronous.

I think the performance edge that Firewire 400 has over USB 2.0 has to do more with USB having a host/peripheral scheme, where the low level protocol operations rely on the host processor; while FW is more peer-to-peer oriented, so those I/O operations are carried by the device controller and the host controller; as a result, data transfer rates are not severely affected by the CPU load or speed in the latter case. Of course, this scheme makes Firewire chipsets more expensive.

Re:Great.

[Agripa]

Err the PCI-X bus is only rated at 1.06 GB/s (or 4.3 GB/s for PCI-X 2.0) so what exactly is the point?

Are you confusing PCI-X (PCI Extended) with PCIe (PCI Express)?

PCI-X is typically operates from 4.3 Gb/s (64 bits at 66 MHz) to 8.5 Gb/s (64 bits at 133 MHz) and is normally only used in servers or workstations.

PCIe starts out at 2 Gb/s (twice the speed of PCI) and easily expands to handle higher transfer rates by using multiple links. x4 and x16 links are common.

Re:Great.

[DaveWick79]

Flash may be slow to write, but it is still volumes faster than a hard disk write.

tenfold or 10 times

[Anonymous Coward]

If it is currently 480mbps, a tenfold increase would take it to 480*2^10 = 480 * 1024 =~ 480gbps.

Re:Honest Question

[skives]

Ethernet is all signal, no power. You couldn't charge your iPod or iPhone with your sync cable, nor would your mouse or keyboard work without batteries.

It's a nice idea though... Ethernet along with the usual IP and TCP/UDP slapped on top are very mature standards that are easy to implement and troubleshoot. The recent debacle rolling out huge SAN networks at my company (for Sun servers) has me rethinking the viability of NAS just for those reasons.

Re:Cable?

[compro01]

the main problem is propertory connectors. there are a grand total of 4 USB connectors. A, B, mini-b, micro-b.

past that, there're 18 bazillion kinds of "USB" connectors made by stupid companies.

What about PoE

[Anonymous Coward]

Ethernet is all signal, no power. You couldn't charge your iPod or iPhone with your sync cable, nor would your mouse or keyboard work without batteries.

That would be "Power over Ethernet".

http://en.wikipedia.org/wiki/Power_over_ethernet [wikipedia.org]

USB is simplex, FireWire is Duplex

[IvyKing]

In high speed mode, the same pair of wires is used for transmit and receive - FireWire has separate pairs.

Wrong

[RingDev]

USB 3 is 4.8 Gigabits per second.
That's .6 Gigabytes per second. .6GB/sec = 600MB/s. .9TB = 900,000MB.
900,000/600 = 1500 seconds

Assuming the drive itself is empty, formated and has a sequential write speed to keep up, you are looking at 25 minutes to fill a .9TB drive.

-Rick

Re:Honest Question

[smellsofbikes]

A bunch of other people have chimed in with parts of the answer already, and I thought I'd add some more.
Obviously each manufacturer wants you to use their standard and buy their hardware.
Different implementations came through at different times, and have different amounts of software/hardware overheads, and try to do different things.
RS232 has almost no necessary software overhead -- you do any and all the work with code you write. USB has *quite* a bit of software overhead to deal with device identification, and Ethernet has *enormous* amounts of overhead. In most small systems you have to buy an Ethernet software stack separately from the OS you're using.
USB tries to provide power. People are trying to glue power into Ethernet although it hasn't yet taken off.
People keep going off in odd wireless directions.
The fundamental problem, I think, is that there are several different connectivity needs and manufacturers are trying to get you to buy their solution to what they think are the most important needs. What you're asking for is something good for the industry in the long term, and that's not really in the direct interest of any particular company, so nobody's building anything for it.
The military embedded market seems to be moving towards gigabit or 10-gig fiber ethernet for all their interboard communications, but fiber has its own problems, and I'm not sure it's the right thing for a USB key you're carrying in your pocket all the time.
I believe that the SCSI module in linux handles firewire and USB, so from that standpoint it looks like it's a start towards universal communications, except for Ethernet. (Even though old SCSI is nothing like serial: it's the the ultimate expression of parallel communications, with some similarities to the old HP/GPIB parallel communication standard that's still used in for test communication but used to be a hard drive standard.) I have no idea what Windows does.

0.9 TB / 4.8 Gb/s = 1500 seconds

[Anonymous Coward]

WTF, a chain of "Exactly" and "Indeed", and no one realizing that 1.5 seconds is wrong by a factor of 1000?

Re:speed over fiber vs speed of electrons over cop

[Shotgun]

GE is limited due to the speed of light and the way the ethernet protocol works. A sender has to stop sending if it senses someone else talking on the same line. In order to do this, it has to detect the collision before it finishes sending. If the line is to long, a sender at each end will be able to get an entire packet out before being able to sense the first bits from the other end. Ugly things happen then. Google "CDMA-CS" if you really want to know more about what limits the length of ethernet.

EM interference is handled by the twisted pair. A pulse of EM energy will cut across the signal and ground wire at the same time. The reciever senses the difference in the voltage levels across the pair, so if you effect both at the same time, the reciever doesn't know (or care).

Fiber has it's own host of problems, but for these short distance and relatively low data rates (for optics) they can use lossy plastic cables with 1/8" headers and just pump LED power to make up for the loss.

Re:Cable?

[ddstreet]

> like if you plug a usb 1.X device onto a usb 2.0 bus, then everything slows to usb 1.X. IINM...

This is wrong, if you plug a USB 1.1 device into a USB 2.0 "bus" then it does NOT slow everything down. Specifically there are 2 cases:

1. You plug a USB 1.1 device directly into your computer (i.e. directly into the "host controller"). In this case, the USB 2.0 host controller (technically a EHCI chip) does NOT talk to your device. Instead, the EHCI chip has one or more USB 1.1 host controller chips (technically either a OHCI or UHCI chip, and called a "companion" chip when inside a EHCI chip) and your USB 1.1 device is connected electrically to that controller. You device is not on the USB 2.0 (EHCI) bus.

2. You plug a USB 1.1 device into a USB 2.0 hub. In this case, the USB 2.0 hub creates a complete USB 1.1 environment specifically for your device. On the host-facing side of the USB 2.0 hub, all communication continues to take place at USB 2.0 (i.e. 480Mbps) speeds. When the host wants to talk to your USB 1.1 device, it uses what is called "split transactions" to talk to it. Basically (I'm simplifying), this involves sending a "start" packet to the USB 2.0 hub. Then, the USB 2.0 (EHCI) controller goes on to do other things, while the USB 2.0 hub initiates the transfer to your device at USB 1.1 speeds. And data transferred from the USB 1.1 device is stored temporarily in the USB 2.0 hub. Eventually the USB 2.0 (EHCI) host sends a "finish" packet to the USB 2.0 hub. If the USB 1.1 transation finished, the USB 2.0 hub responds successfully (either with the incoming data or a "ack" that the outgoing data was sent) which completes the transation.

(There is also a combination case of those, where the EHCI chip does not contain a "companion" USB 1.1 chip, but instead contains an internal USB 2.0 partial hub - the "transaction translator" part - that handles talking to USB 1.1 devices. For bus usage purposes, this is effectively the same as using an external USB 2.0 hub, since the USB 1.1 devices do not appear on the USB 2.0 bus.)

Re:Great.

[Bassman59]

1.5 seconds if you all of your components were fast enough. The drive won't be.

Exactly. There's still the slowdown associated with the mechanical aspects of the hard drive -- spin rate (RPM), average seek time (ms), etc. On top of that, most hard drive controllers are limited by the technology they use. For instance, a SATA hard drive, even plugged into a USB 2 or 3 port, is limited to 150 MB/s -- but, that's burst speed, not sustained transfer rate.

you people are all missing the point.

It's obvious that the 4.8 Gbps link is faster than the device ... but recall that all USB devices on a port share the bandwidth. A faster link will allow a lot more devices to simultaneously transfer data at their maximum-possible speed.

One example: You'll be able to put a multichannel audio I/O device and hard disks on the same bus without worrying about dropouts, etc.

As for "Isn't 4.8 Gbps faster than the computer can handle?" That's true, and it's already the case when you have Gigabit Ethernet plugged into 33 MHz/32-bit PCI. But remember that 4.8 GBps is the wire speed, not the data rate, which will ultimately be controlled by how fast the host computer can get data onto and off of the bus.