Comms chips

Intel Revisits Network Processors

Intel Corp. (Nasdaq: INTC) might be ready for another crack at the network processor market and at telecom-grade digital signal processors (DSPs), Light Reading has learned.

Notice the word might. Intel isn't announcing anything yet; even slideware wouldn't emerge from the company until at least next year, says Stephen Price, a marketing director with Intel's embedded and communications group. And Price warns that plans do change.

Still, it's a reversal for Intel to even think about such a move. The company's IXP network processors were a top seller, but as the telecom boom of 1999 faded, so did Intel's interest. The high-end branch of the devices is now in the hands of Netronome. (See Will Intel Trash Telecom?, Intel Licenses Netronome, Netronome Systems Inc., and Netronome Reigniting Intel's IXP.)

Now, it appears Intel's Xeon line of microprocessors has reached a level where the company has started thinking about them as the basis for a telecom revival.

"Our products are getting more aligned with telecom than what I've seen from Intel in the past," Price says. "We're looking for the share of wallet in some of these things done by DSPs and network processors."

(On the DSP front, he's talking about large, high-performance DSPs. Not the wussy kind that go into greeting cards. And, please, turn the volume WAY up when you click that link.)

What's crucial to the idea is that Intel wouldn't just "throw cycles" at the problem. That term refers to using a really big general-purpose processor to do a job that's meant for a smaller, specialized processor. You get the job done, but inefficiently; the general-purpose chip needs more brainpower to do it.

Price wouldn't give details of Intel's plans, but he suggests there's an ideal mix to be had of programmable chips (FPGAs), traditional DSPs, and Intel Architecture chips.

"You can imagine what we could do with a combination of big cores and small cores, to deal with anything that's lookaside or in-line, whether it's Layer 1 through 3, or up to Layer 7."

The company has a bit of a head start on the DSP front, too. Some customers are already offloading DSP work to Intel chips, Price says.

It's interesting to note that Cavium Inc. (Nasdaq: CAVM) -- which competes with companies like Intel, Freescale Semiconductor Inc. , LSI Corp. (NYSE: LSI), and RMI Corp. for general-purpose sockets -- has similarly said its chips have gotten sophisticated enough to substitute for network processors. But Cavium says it's not interested in the network processor market. (See Cavium Sprouts More Cores.)

Meanwile, Intel's Xeon chips are finding their way into plenty of telecom applications, Price says. The chips are used as control-plane processors in telecom equipment, of course. They're also adept at pattern matching and encryption, making them suitable for security appliances or for Kasumi or Snow encryption in wireless networks.

Intel also believes WiMax and Long Term Evolution (LTE) base stations will need the kind of processing that the Intel Architecture provides. (Base stations for 3G and earlier, not so much.) Price said Intel would be giving service providers some WiMax and/or LTE demonstrations, in private, during next week's Intel Developer Forum, although he wouldn't give details about what those products are.

— Craig Matsumoto, West Coast Editor, Light Reading

paolo.franzoi 12/5/2012 | 3:56:01 PM
re: Intel Revisits Network Processors


I think I have posted about this in the past but it seems to me that they jump on apparent growth waves.  In the past, these waves were not as profitable as their core business.  Thus, these add-ons were eventually chucked out.  And then they think they are missing a growth wave (rinse/repeat).

First thing I remember Intel abandoning was the 8085.  I know there were probably earlier things, but dumping that was a huge impact on an "Intel" based shop.




Pete Baldwin 12/5/2012 | 3:56:01 PM
re: Intel Revisits Network Processors

These things run in cycles, don't they?  Intel gets the itch to expand beyond the traditional x86 world, puts money into R&D, pulls back (or says it was just 'seeding the industry') when the economy toughens up...

The network processor idea sounds like something that could end up stifled quickly through no fault of its own. That is, its half-life might shorten if there's just the right (wrong) management change, that sort of thing.  They might give DSPs more of a chance -- but that's just a guess.

Either way, we might end up adding them (back) to the list of things Intel tried and discarded.  Speaking of which, we have an ATCA conference coming up this week at ESC.  :)

Sign In