IA32 is pushing elephants up steep hills - Bob Colwell, former chief architect at Intel
EVERY PC these days has a Gigabit Ethernet connection - many high-end ones have two in fact. Nvidia chipset-based entries, as well as Intel-based workstations, also have TCP/IP offload to help improve the throughput without taxing the processors.
What about 10 Gigabit Ethernet (10GE), the next standard for manhandling all those TCP/IP packets over the networks? Well, for now it is stuck in closed-door supercomputing labs of various governments, univesities and alike "institutions of highly guarded secrets", and in a couple of datacentres, here and there.
Why? It is simply too expensive - a quick check with, say, Cisco, Force10 or alike will give you an idea of a cost of the 10GE infrastructure, say NIC card, switch and CX4 standard cabling - close to two kiloquid per port, or a bit cheaper in the US of A, around US$ 2,000 to 3,000.
That IS expensive - but not anymore. Fulcrum, a recent fabless Californian high-end network switch IC outfit - formed by an ex-Caltech gang - seems to have a solution. One that could, in a year or two, bring 10GE closer to the ultra high end PCs and workstations too.
Of course, it still has to start at the current datacenter level, and fill that up first before moving beyond. Fulcrum's FX4000 single-chip 24-port full-bandwidth 10GE Ethernet switch is simple, inexpensive and very fast, at just 200 ns port to port hop speed in Layer 2, or 300 ns in Layer 3. It also does nifty things like preventing DOS (Denial Of Service) attacks and flood control, all in hardware.
The reduced latency is important to supercomputing buffs, where, combined with a fast 10GE adapter like those from Myricom, it can do a message passing write in just about 2 microseconds now, from one system to another. Therefore, it matches the speed of InfiniBand, a wannabe "standard" which for a long time attempted to be a 'jack of all trades' from networking to storage and clustering.
The integration lets you have up to 48 10GE ports in one slim 1.75-inch high switch, but also pack it inexpensively enough even for server or workstation benchmarking labs to have one or two. Example: Quadrics TG201 24-port switch from Bristol, some 80 miles from The Inq office, at just over US$ 7,000 - good for some university cluster play, or Arastra 7100 48-port unit at around US$ 19,000. You can also have autosense 1GE / 10 GE ports with common cabling in the latter case.
Now, the huge bandwidth and low latency could enable some futuristic high end
workstation and PC use models too. For instance, a pair of "multimedia servers"
with dedicated 10GE connection each, linking to a home network of 1GE connected
PCs, all through a single switch, able to stream uncompressed HD or lossless 4K
cinema video to all of them, at full guaranteed frame rates without any drops
due to the low latency.
On a slightly larger scale, Fulcrum has a big version of this, called "Vegas ", a combination of 10GE and 1GE ports in a single switch, up to 48 ports. You can federate the chips to go up to 144 ports in the near-future products, too.
I'd prefer this to any 802.11n home or office-wide connection, not just because of added speed; you avoid the radiation exposure effects of a zillion access points around you...
Being a small company fed by the venture capital - one of the backers is Andy Bertholsheim, the Sun system guru - Fulcrum has to be concerned with its competitors getting focused on it. For now, they seem to have an advantage over the competitors with this gadget.
However, other makers like Fujitsu and Broadcom will surely follow with fast and more affordable 10GE switch circuitry - how about Intel doing something to get out a fast, affordable 10GE PCI-E adapter for PCs and workstations? µ
When talking about 10G Ethernet connection, everyones forgets that it's almost useless. 

why? becuase of the hard drivers of course. SATA, at the maximum, can operate at the speed of 3GB/s. and I haven't heard of the hard drive who can even manage to transfer at that speed. so for 10G Ethernet connection to be useful, you need a raid array of a couple hard drivers, and that is not so inexpensive.

The bottom line is this: we need a faster storage solution :)
Yep, the problem with wireless is that it obviously and painfully does fry your head.

Networking hardware is currently in the same place that IBM-compatible and Apple PCs were at when they first became things people bought for the home - ie, you can only buy them pre-built.

It's somewhat concerning that 10 gig Ethernet is seen as being still out of reach for home networking though - in China and South Korea and Japan, and in some European countries (not the one I live in of course), fibre-optics to the home, for internet connections, already exist and have been around for a while.

Fiber's way faster than 10 gig, and of course more secure - it's not exactly easy to spy on a fibre link even if you manage to cut into the cable without disrupting the transmission.
The secret sauce behind Fulcrum's performance is their use of asynchronous VLSI technology... this gives low power and low latency. It may be hard for established companies to match their price and performance using standard design methods.
Speaking as a radio amateur, the 1 Watt limit on emission standards in the 2.4Ghz band is hardly something to worry about. My triband HT (handy talkie) outputs 5 Watts, and is placed much closer to my head when I'm using it. The human body is also much more susceptible to RF radiation on frequecies that my radio operates on.

If you have a window on the side of your computer case, or if you leave the case off all together, you're probably emitting more RF than a typical wifi card of 200mW or less.

One thing I will agree on is that unless you're surfing the web, wifi really is teh sux0rs. :)
Here, check these out - 

http://www.itweek.co.uk/vnunet/news/2204531/cisco-sets-40gbps-broadband

http://www.itweek.co.uk/vnunet/news/2194238/granny-gets-40gbps-internet

And that's not even pushing fibre tech to anywhere near it's current limits.

And RAIDS are very commonplace in mid-range rigs - lots of well-priced motherboards come with 0, 1, and 10 onboard, and many have 5 also. Large capacity drives with decent caches and seek / read times don't cost that much.
eg - 3 x 500 gig drives (7200rpm, 3 platters on each, 16 cache) is just under £200, 180 quid in many places.
Plus there's already solid state hard drives available too.

Plus it's not an issue when it comes to downloading video & audio and watching TV, and with online games there's other considerations that assist in the hard drive usage speed - you need enough RAM (which keeps getting cheaper and faster) to be holding the data in real-time as it writes to the drive, and obviously good video card(s) (and even physics cards) so the GPUs take that load off the CPUs, and there's already network cards too (& some onboard versions of that) that take the network traffic tasks away from the CPUs.
arcanes, nobody's arguing that a faster storage solution wouldn't be nice, but that's not the thing that 10GbE is useful (read: necessary) for. Mainly it's streaming video. Not just video files, mind you, but rather uncompressed video output streams as the author mentions. 

Two words: remote interface! Put your computer over here, watch your movies and play games and crap on your awesome, expensive flatscreen that I don't have, over -there-! Now flatscreens can enjoy their solitude- the PC is akin to a CRT display, and is noisier to boot, so isn't it a rather mundane leap of logic to move it somewhere else and keep only the display output in the living room?

Coward, what do you mean by a-sync VLSI? More specifically, how is it used to advantage here... it's a full-duplex connection, no? Or at least can go the same speed both ways? I just don't see how the async part helps them out- of course the VLSI does; without it there'd be a stack of tubes the size of a large house to accomplish the same thing. Also, FYI it's usually called ULSI these days. Why the continuous changing is necessary, I know not.
What about live video or the ability to run large back end programs through terminals which up till now sucked because the connection speed was slower than my mother in walgreens with a stack of coupons.
It will take more than just a few consumer drives raided together to get you to 10GE! Even a NetApp 3070 $400K filer can't fully saturate a 10GE! Trust me on this, I have one! Take 10 300GB 10K SAS drives and RAID them together in RAID-5 using a high end PERC5/e card, and you'll get 200MB/s (1.6gbit). 10GE just isn't even useful in the desktop environment yet. It can only just barely be used in the data center (mostly for link aggregation).
There is no Ethernet solution capable of 2 microsecond end to end latency. While switches may appoach 200 ns switching latency, the best RDMA (iWARP) numbers over 10 Gigabit Ethernet are around 7 microseconds for the MPI protocol.

Myrinet may get better, but Myrinet requires Myrinet switches, not Ethernet switches, plus, published Myrinet 10G numbers are based on back to back server connections (no switch).

InfiniBand (which is a standard, not a "wanabe" standard), current shows MPI latencies of about 1.2 microseconds using the latest Mellanox ConnectX host channel adapters connected with a double data rate switch.

As much as Fulcrum and others like to think they will attack the same HPC clustering market InfiniBand current addresses, the real market for low-latency Ethernet (using iWARP) will be the much larger market where 10 usec to 15 usec latency is good enough, not the sub 5 usec latency market, which will remain InfiniBand.
Well if there's 1000MB per GB, and it's easy enough to get SCSIs with 80MB STRs, you're only needing about 15 of them to easily get 1GB in transfer time. 15 x 40 HDDs is nothing in terms of a datacenter of today.
And given this is about live streaming and gaming - not saving anything - there's no immediate need for backup (redundant) disks, a RAID 0 would do fine thus lowering price. Any saves could be done later, same with any webpage or online content that you want to keep.

It seems the bus systems need to be speeded up to accomodate existing hard drive speeds overall anyway.

Here's an HDD that uses DDR1, it'd need an UPSU for data security (note it loses content if all power is cut off, as with any RAM), and only goes up to 16GB, but picture it done with low latency DDR3 and at larger capacities (there's loads of server motherboards using various RAM standards that take 64GB of RAM for example):

http://www.hyperossystems.co.uk/

And then there's this:

http://colossalstorage.net/home_diskdrive.htm

Call me old-fashioned, but it don't look like they're worried about anyone stealing their ideas given the details gone into for the technology there. So they must have it pretty close to completion at least, all patented and so forth. In fact, if I were to place a wager on it I'd say such tech is being held back purposely though not by the folks wanting to roll it out of course.

Don't get me wrong, best I can afford just now would be a set-up with 3 drives max. But this whole worlds been set-up wrongly from the outset, the reason we're so backward technologically is the same reason we're so backward with doing technology correctly - renewable and non-polluting etc. The economy is a big sham that serves a greedy few and it's based on war and destruction, not progress and good-living.