View Single Post
  #7  
Old January 1st 11, 04:47 AM posted to microsoft.public.windowsxp.hardware
Paul
external usenet poster
 
Posts: 18,275
Default How many USB ports needed and supported?

James Silverton wrote:


The only problem about installing further USB ports is whether or not
the mother board will supply sufficient current to run plugged-in
devices. I have also had interference problems with externally powered
USB hubs.

As a further question, does anyone know if any standard makers supply
USB 3.0 ports?


The motherboard has fuses installed at a rate of one 1.0+ amp fuse for
every two ports. On a 12 port motherboard, that is a total of 6 fuses.
They would be connected to +5VSB (a mistake).

In the past (perhaps three years ago), you could get motherboards with
USBPWRxx jumper plugs on them. Such a setup, allowed all the USB ports
to be run from +5V or +5VSB. The former is used, for raw power. The
latter is selected, for those ports you want to serve as a wake up
function. For example, on one of those older motherboards, keyboard and
mouse would be run from +5VSB, while the other USB ports would be
jumpered to +5V. That gives the best of both worlds. Wake-up capability
(press key on keyboard, or move the mouse, and the computer starts),
as well as giving plenty of power to the other ports. By providing
jumper plugs, one jumper per two port stack, the user had the ability
to select the right mix of +5VSB and +5V supply. By putting most of the
ports on the +5V rail, there is plenty of power. On an older
motherboard, I can run all the ports at max current.

The +5VSB rail of the power supply, has limited current. The power supply
in the computer I'm typing on, has +5VSB @ 3 amps, which is enough for
six ports at max current. If I was trying to run a dozen 2.5" USB external
hard drives, the power supply would cut out.

The motherboard really needs those old jumper blocks. The other power
path, the +5V supply, can carry about 24 amps using the wire harness
on the power supply. So that power source, could handle a dozen hard
drives at 0.5 amps each. My power supply can supply 30 amps on +5V,
so the wire harness is the rate-limiting step. And 24 amps would
be enough for 48 USB ports.

Some architect has decided to needlessly saddle +5VSB with all the
USB loads. The +5V rail has enough power to do the job, but then,
ports powered by +5V, wouldn't support waking. So to support any
USB port, being able to wake the computer, the power distribution
scheme is less than ideal.

I expect a Dell/HP/Acer would be similar. Jumper blocks removed, to
save a few pennies. And ports run from +5VSB. So there will be
limits as to the total power you can draw, and the rating on the
label of the power supply, will give the necessary details.

There are "other loads" on the motherboard, on +5VSB. My power supply
can provide 3 amps max, and 1 amp of that, is used by standby requirements
for my LAN chip, to run my keyboard (PS/2) and so on. So I might have
2 amps left "to play with". That would be enough for 4 max current USB2
loads.

Most of my USB2 devices are low power. I don't use 2.5" drives, and my
external 3.5" USB has its own power supply. My 3.5" drive would only draw a
couple milliamps from the bus connection. So in my case, the 2 amps total
would go a long way. But I can easily see, how a different set of
peripherals (lots of bus powered 2.5" USB drives), would crush that
power rail.

*******

USB3 ports, are provided solely by a NEC USB3 chip. NEC was also one
of the first companies to make the USB2 chip (before it was a standard
part of every Southbridge chip). So history is repeating itself, in
that NEC is first with a chip for the newest USB3 standard. For some
reason, both Intel and AMD/ATI, have blown it.

The NEC chip, to run at decent speed, connects to a PCI Exprees x1 Rev2
port at 500MB/sec. The chip has two, USB3 ports. You couldn't run
both ports flat out, if expecting to get record setting transfer rates.
I think the theoretical best you can do, is somewhere around 336MB/sec
(using a new protocol over USB3). Two ports couldn't transfer in the
same direction at that rate, at the same time, using the NEC chip (that
would be 672MB/sec though a 500MB/sec path). But if you were doing a
disk to disk transfer, all hardware paths are full duplex, and in fact
you can do that. (This drawing shows an unlimited transfer test case.)

PCIE x1 R2 Two
Port 336MB/sec
----- 500MB/sec ----- NEC --------------- to hard drive #2
---- 500MB/sec ------ USB3 336MB/sec
--------------- from hard drive #1

Currently, I'm not aware of a peripheral, that supports that
transfer rate. So the theoretical value, hasn't had a workout yet.
I think somewhere around 200MB/sec is the best so far (using an
SSD drive as the hard drive).

If you buy an Asus or Gigabyte motherboard, and built your own computer,
those come with the NEC chip installed. I think there are even some
motherboards, that have two of those chips installed (just for sheer
excess). It's actually hard to do a good job, due to the difficulty
of making PCIE x1 Rev2 ports using existing chipsets. A lot of motherboards
only have left-over Rev1 PCI Express lanes to work with, which prevents
getting to the theoretical max. That's one of the things I try to check,
when shopping for motherboards. If you look at the Gigabyte motherboard
user manual, it sometimes provides enough info, to determine how
the bus connection is done.

Even a broken USB3 implementation right now, will be able to sustain
the transfer rate of any single disk you buy. So it's not like
this matters much. It really depends on whether you think you'll be
able to afford a 336MB/sec device whenever such a device comes out.
The best hard drive I own, can do 130MB/sec sustained, just to show
how far from max it is.

http://www.nordichardware.com/index....ticle&id=20792

Computers from Dell/HP/Acer/Gateway, aren't likely to have that chip.
The reason is philosophy. Those companies, don't really like to
add a lot of peripheral chips to the motherboard. They usually wait
for companies like Intel, to put the necessary ports right on the
Southbridge. And that hasn't happened (for Intel's business reasons).
The idiots running the show, seem to be betting on Light Peak, to
sell a lot of product. And USB3 is falling by the wayside. Only
enthusiasts, buying their own motherboards and building their own
computers, are getting USB3 right away.

There is another way to do it. You can buy an add-in card for your
computer, and just add the ports to your Dell. Architecturally,
this is about the best of the bunch, as it tries to max the
PCIE bus bandwidth. This is a bridged design, that connects at
PCIe x4 Rev1, to give PCIe x1 Rev2 ports internally ("gear
changer chip"). But the user reviews aren't all good for this thing.
The card is cheap, but read the reviews before buying one.

(Asus U3S6 add-in card - PCI Express x4 connector)
http://www.newegg.com/Product/Produc...82E16813995004

You need an expansion slot, of either x4, x8, or x16 size, to take
that particular card. I have two x16 slots on my motherboard, with
the video card taking one slot. So I could fit that x4 card, in my
second x16 slot and use the card. That would be how I'd add USB3, at the
full rate, to my old motherboard. If you bought Dell, you'd want to
check the slot types, to find the right expansion capability to add
that.

The introduction of USB3, was one of the weirdest hardware intros
I've ever seen. The first thing to show up, were USB3 cables.
No peripherals or chips. But you could buy cabling. Next,
NEC delivered their chip. And to speed introduction, NEC
also contracted an unnamed third party, to build PCI Express x1
USB3 cards, with the NEC chip soldered to them. Many companies
bought those contract manufactured cards, for the initial release
(and rebranded the cards, by placing a sticker on them).
That helped NEC get the chips out there quickly, so they
could make some money on their design. But the funny thing
is, NEC still seems to be the "only game in town". To
the victor, go the spoils.

Paul
Ads