WiFi uber alles

I am playing catch up with Cliffside, an Intel project to put WiFi in the personal area network disclosed at IDF Shanghai. Turns out this work and another effort (stay tuned for a Monday story) has a lot of people excited about WiFi someday becoming the one radio you need to handle everything from mice to linking to the Net.

One of my fave wireless analysts, Craig Mathias, claims the time has come for WiFi PANs. "In a couple years a WiFi PAN business could be very successful," Mathias told me on the phone today. It won't kill Bluetooth--nothing dies in high tech--but it could stunt its long term growth worse than a bad cigarette habit.

Intel claims its technology can connect up to eight WiFi devices to a notebook on a PAN while the computer is on a WiFi LAN. The technology could improve the quality of media streamed between a computer and a TV or other display because it eliminates the latency of going through an access point.

The Cliffside program uses a modified Intel WiFi chip with additional buffers to rapidly switch between PAN and LAN modes and is expected to ship within 12 months, said Gary Martz, a marketing manager for the program. He said use will probably be limited to consumer notebooks for the first year or so.

"Corporate IT managers will probably have a heart attack, perceiving this as enabling notebooks to be a bunch of rouge access points that impact performance of their WiFi networks," Martz said.

Mathias was animated in his enthusiasm for WiFi. "WiFi's success is unquestionable. It will be in everything. Twenty years from now people will still be talking about WiFi."

Good, we tech reporters need the job security ;-)

Ethernet's puppy

I haven't had my eye on the ball of the high-speed Ethernet standards effort over at the IEEE 802.3ba for awhile, so I talked to the chair of the committee, John D'Ambrosia late last week for an update. This is big complex stuff about the future of Ethernet, the data center and converged telecoms at mind-bending 40 and 100G speeds.

John reports all goes well with technical progress at the last meeting, as many as 200 expected at the next meeting and appropriate debates on the floor. I haven't heard much from the committee at large about this effort, so if you have some observations or opinions on this big leap forward post a comment or drop me a line at rbmerrit@cmp.com.

While this standard is gestating, John has a new puppy at home (pictured while snoozing) that is also growing fast. John says it's possible the .3ba work could be in an early draft stage by March and a late draft in November. By that time little Buster D'Ambrosia will be chomping down dog food at a 40 or maybe 100G rate.

Quick predictions on graphics

I spent a couple days researching GDDR5 for an EE Times story last week. From what I learned, here are a few quick and free from-the-hip predictions. Take them for what they are worth.

--AMD will be the first to use GDDR5 on a new graphics controller to be announced soon.
--Archrival Nvidia will wait as long as possible before moving to GDDR5, opting to use the more economical GDDR3 at 1.3 GHz+ as long as possible
--Intel will use GDDR5 on its much-anticipated Larrabee controller when it debuts probably in late 2009, but the part won’t make that big of a splash in the market in its first generation.
--Rambus will struggle to get traction with XDR in HDTVs and elsewhere, crossing its fingers for better luck in next-generation videogame consoles with a second or third generation technology.

Think I'm wrong? Share a prediction of your own here or at rbmerrit@cmp.com. What to learn more? Check out this white paper.

China rising

You get interesting glimpses into where a company is at trolling the job boards. I saw a recent ad for Hua Wei on the SI-List for an engineer that indictaed where this company is on its climb from a copier of Cisco products to an innovator of modest capabilities in its own right.

The ad called for an SI designer who ideally has "experience in taping out multi-million gate CMOS ASICs in 130nm technology and below." They are willing to take someone with no more than a bachelor's degree and two-years industry experience to handle "evaluation and choice of 3rd party IO buffer and SerDes IP for ASIC design."

Clearly, they are not at the exalted level of Cisco's 90nm Quantum Flow Processor or even its FCoE work, but few ASIC designers on the globe could pull off a design like that. Nevertheless, the China's answer to Cisco is clearly on the rise, and I'd like to—and expect to--hear more from them.

8G Fibre Channel heats up

In the relatively slow and steady pace of the Fibre Channel world, competition is starting to heat up around the transition to 8 Gbit/s products. Storage switching giant Brocade rolled out its portfolio of 8G switches and server cards today.

Emulex was quick to point out it has been there, done that with server cards already qualified by EMC. Indeed both Emulex and QLogic were early to the 8G party with cards they rolled last summer.

Despite a little bit of temperatures rising, I recall what Renato Recio of IBM told me during an interview for a recent story about Fibre Channel over Ethernet. For users that what high performance and need to move this year FC is the safe bet, but for 2009 the FCoE community should have its ducks al in a row to be a solid competitor.

Ethernet gets green

There's plenty of work to do at the IEEE 802.3az, but Intel's proposal for a low power idle version of Ethernet has already got the nod to be part of the group's spec for 100 Mbit and Gbit chips. The next big debate will be whether it or a Broadcom proposal for a so-called subset PHY wins out for 10 Gbit chips and for backplane Ethernet.

Next week's meeting is the final call for proposals, so if you have a hot idea for green Ethernet better find the meeting and come with Powerpoint in hand or forever hold your peace. I'd be happy to host a debate on the trade off between these two approaches if anyone wants to take one side or the other in comment posting here.

The Conexant/CopperGate connection

No big surprise that Conexant sold off its powerline networking group today. Linking the dots on its future, it's not a big shocker HomePNA specialist CopperGate Communications was the buyer.

The Multimedia over Coax spec is Conexant's current direction in home nets these days. A Conexant exec told me at the MoCA conference last fall that Conexant plans to ship discrete MoCA RF and baseband chips in the second half of 2008, then follow them up with chips that integrate MoCA support as a block on its silicon for DSL, MPEG and passive optical network chips for set-tops, routers and optical network terminals.

Conexant and Broadcom both joined the MoCA board recently, telegraphing plans for MoCA silicon. Conexant had already given up on earlier plans to make HomePlug AV chips, believing the technology will have difficulty keeping pace with the bandwidth needs of digital media.

For its part, CopperGate is very bullish on the ITU's G.hn standard as part of its road to the future. CopperGate just announced it has delivered five million chips for HomePNA over phone and coax lines to date. It bought Connexant's powerline group so it could add to its mix that technology which European carriers demand.

Looking to the future, CopperGate foresees the day when a completed G.hn could allow it to deliver a chip that handles all three wired media in a standard way. The company also plans to offer chips that support G.hn as well as the version 4.0 of HomePNA now in progress. One chip, all home net markets--that's the Holy Grail.

I wonder why Conexant didn't see that broader win, hang on to its power line group and prepare for the same single-chip future.

RapidIO fans out

Tom Cox pointed out to me that I missed a milestone back in February when RMI Corp. (formerly Raza Microelectronics) announced a processor for wireless base stations using a native serial RapidIO interface. It was the first MIPS chip with the interface that has previously been used in PowerPCs and DSPs.

No big surprise that anything linking into the wireless base station might go RIO, given the traction the group has gotten in DSP farms there with Ericsson, TI and others. But will the MIPS win expand beyond RMI?

Cox, executive director of the RIO trade group, suggests the right dynamics are in place. Cost reduction, the need for better communications links and the rise of multicore processors will drive out FPGA-based bridges in use today between MIPS chips and RIO farms, he said.

The RIO group has gotten several new members recently including Nortel and Qualcomm. No word on exactly what these folks are up to yet. If you have some insights, post a comment or drop me a line at rbmerrit@cmp.com

Waiting on Home Net 2010

Jockeying for position in the home network circa 2010 will be a major consumer sport with the HomeGrid Forum jumping into the fray. The players all will want to position themselves as the best solution for everyone from carriers to OEMs to consumers.

So far MoCA, HomePlug and HomePNA have been pretty quiet about their plans to leap toward the 400+ Mbits/s carriers are calling for in the next year or so. But everyone has heard that call including the HomeGrid folks who want to goose the G.hn effort to deliver that over coax as well as a reach goal of Gbit links when possible.

Writing an analysis of the latest news revealed a few interesting wrinkles. For example, the wired (MoCA, Powerline and HomePNA crew) represent just 19 percent of today's home nets which are dominated by WiFi and wired Ethernet. And most of HomePNA's installs are on coax, not phone lines.

To get one overarching standard will require a tussle between the big vendors who want a unified market and the small vendors with silicon skin in the game. Already Pulse~Link is suggesting G.hn should re-think its choice of OFDM (which it does not use), while HomePlug folk praise the group for the choice (because they made it, too).

Meanwhile the powerline folks at IEEE 1901 have yet to be able to muster support for a confirming vote on a HomePlug/Panasonic home net proposal. The vote was put off at meetings in October, December and March, but may come up in July. Not a good sign for consensus building in this community.