Re: How could native USB 3.0 in new Intel CPUs help?

Workbug wrote:
According to what I read from the link below, the new Intel CPU
doesn't have native USB 3.0 support. But is native USB 3.0 support
important? How does it matter? How much would it improve over the
current USB 3.0 speed?

USB3 is defined by a standard. NEC built a chip according to the
standard, and makes it available at a relatively low price. It
has allowed the manufacture of $25 add-in cards for computers.

One of the participants in the standard (Intel), was accused of
being "ahead of the curve", by playing a key role in defining
the standard, and as a result, of having an unfair advantage.

The funny part now, is Intel is slowing the propagation of USB3
for their own business purposes. It will not stop people from
enjoying USB3 performance though, due to the efforts of
NEC to propagate their design (availability of PCI Express
plugin cards for desktops).

This article yesterday, made me laugh.

"Ravencraft said. It takes "a minimum of two years if not more"
for a company such as Intel to build USB 3 support into its
the chipsets"

All we can do, is point at the NEC efforts and successes, and just
laugh at such tripe.

I think it is interesting, what's in Wikipedia on the subject.

"USB 3.0

The USB 3.0 Promoter Group announced on 17 November 2008, that
version 3.0 of the specification had been completed and had made
the transition to the USB Implementers Forum (USB-IF), the managing
body of USB specifications. This move effectively opened the specification
to hardware developers for implementation in future products.

The first USB 3.0 consumer products were announced and shipped
by Buffalo Technology in November 2009, while the first certified
USB 3.0 consumer products were announced 5 January 2010, at the
Las Vegas Consumer Electronics Show (CES), including two motherboards
by ASUS and Gigabyte Technology. Manufacturers of USB 3.0 host
controllers includes, but are not limited to, Renesas/NEC Electronics,
Fresco Logic, Asmedia, Etron, VIA Labs and Texas Instruments.
As of November 2010, Renesas is the only company to have passed USB-IF
certification, although motherboards for Intel's Sandy Bridge processors
have been seen with Asmedia and Etron host controllers.

...AMD is working with Renesas to add its USB 3.0 implementation
into its chipsets for its 2011 platforms."

What the last statement means is, AMD will potentially buy an IP block
(intellectual property logic design) from NEC, and install it in their chipset.
That will leave Intel to come up with a story. Maybe a guy with a shiny head,
will make excuses for them :-)


With regard to your performance question, as long as there are no bottlenecks
between the USB3 logic block, and the rest of the system, you're getting
the full performance. The NEC chip requires a PCI Express x1 (one lane) interface
of the Revision 2 variety (500MB/sec). From the Wikipedia article:


A new feature is the "SuperSpeed" bus, which provides a fourth transfer mode
at 5.0 Gbit/s. The raw throughput is 4 Gbit/s and the specification considers
it reasonable to achieve 3.2 Gbit/s (0.4 GB/s or 400 MB/s), or more, after
protocol overhead."

5Gbit/sec is the line rate of the cable used outside the PC. On that cable,
the information is encoded. Once the overhead bits are removed, you get
4Gbit/sec. Dividing that by 8 gives 500MB/sec. That doesn't exactly fit
in a 500MB/sec PCI Express x1 Rev2 lane. Both protocols have further overheads
(packet overhead, like a packet header or a CRC on the end), which has to be taken
into account. I don't have info on how much overhead a USB3 packet has, versus
the overhead of a PCI Express packet.

(A poor article, on a particular line rate encoding - they don't explain
why you need this. 8B10B encoding has existed for quite a while. You
send 10 bits down the line, and 8 bits come out after decoding. That's why
the 5Gbit/sec line, delivers 4Gbit/sec inside the chip.)

Now, of the (500MB/sec minus packet overhead) bandwidth, there are
still overheads from a packet protocol point of view. Someone did
calculations, to estimate the performance to expect, and with a
UAS driver installed and a USB3 host chip, the rate to expect
is 336MB/sec. No devices (that I've heard of to date), achieve that rate,
and the last result I saw was somewhere in the ~200MB/sec range. You would
need to connect a USB3 peripheral to a SATA III SSD, to get close
to the limit. (Really, most people would be perfectly happy,
even with ~200MB/sec, for things like file backups, compared to the
30MB/sec they were getting on USB2.)

There are devices, demanding a USB3 host chip with a full 500MB/sec interface
on it. The software for this device, for example, does a bandwidth
check, before attempting to do full rate transfers. (It can
tell the difference, between a NEC USB3 Host controller on a
PCI Express x1 Rev2 500MB/sec interfaces, versus a 250MB/sec Rev1

"Blackmagic intensity shuttle (USB3 video capture)"

Intel was working on something called Light Peak. But something I'd
warn anyone about, is "don't underestimate the difficulties of
using fiber optics". I'm curious what direction Light Peak will
take, and how watered down it'll be at launch. Intel's gamble
presumably, is their new technology, will be intended to leapfrog
USB3, and yet remain as cheap as USB3. Leapfrogging is easy,
being cheap, is not. And would any purpose be served, by going
even faster ? It's likely a technology, ahead of its time
(and application space).