Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.
April 14, 2009
Chipmaker Cavium Inc. (Nasdaq: CAVM) is upgrading its microprocessor line with an entry that, on paper, outdoes the big companies like Freescale Semiconductor Inc. and even Intel Corp. (Nasdaq: INTC).
The Octeon II family, getting announced today, will sport up to 32 cores on a chip -- each "core" being a small processor in its own right. That's a mouthful of processing ability. By comparison, Intel's Xeon line is just getting to eight cores. (See Cavium Launches Octeon II .)
That's decent bragging rights for a company that didn't start out in this market in the first place. Cavium was initially a security-chip company, but it's better known these days for the Octeon lines.
Full 32-core firepower won't be available right away. The first Octeon II chips, formally called the CN63xx family, will sport two to six cores. Earliest samples of those chips are expected to come out by the end of the year.
The presumably bigger CN68xx line will be Cavium's next step, but the company doesn't plan to give away any details until after midyear. Lower-end chips will be on the way, too, aiming for systems that need lower power consumption and low cost but still want all those processor cores.
That's different from Cavium's usual plan. "Previously, they started from the high end and then started lopping stuff off," says Joe Byrne, an analyst with The Linley Group .
The 32-core version of Octeon II will be enough for handling 40-Gbit/s traffic, Khemani says, but Cavium is claiming just 10-Gbit/s throughput for the CN63xx line. (Older Octeon chips already go 20 Gbit/s, but Cavium says the CN63xx will target a more mainstream market than those chips do.)
Octeon is a breed of what are called embedded processors -- a loose term used for nearly any microprocessor that doesn't go into a computer or server. Freescale is one of the top companies in this market; so is Intel, although it's tended to put more focus on PCs than on the embedded market.
The scope of the embedded market is immense, and that's partly why Cavium is able to make such bold claims against Intel and Freescale.
"In that regard, Cavium is ahead of either of those two companies, but Cavium is focused," Byrne says. "It's one of the benefits of targeting a specific workload or type of application."
Another benefit: In addition to packing more cores, Cavium is able to lace its chips with interfaces specific to networking or circuitry designed to help parse packets, Byrne says.
It's possible for Cavium to compete with network processor companies, such as Bay Microsystems Inc. , EZchip Technologies Ltd. (Nasdaq: EZCH), and Xelerated Inc. In fact, Cavium points out that Octeon II can be used as a combined control-plane and data-plane processor, saving space on a line card.
But network processors tend to chase a higher-end carrier market that Cavium doesn't consider a priority.
"In comparison to the market we are chasing, it's small," says Rajiv Khemani, vice president of Cavium's networking and communications division. "We're covering from the low end all the way up to the 40 Gbit/s range."
Likewise, Khemani says Cavium isn't interested in making any 100 Gbit/s proclamations yet, because the market is so relatively small.
Cavium reported $87 million in revenues in 2008; Byrne estimates about $50 million of that came from Octeon. By comparison, another focused processor company, RMI Corp. , tallied about $44 million in revenues last year, by Linley Group estimates.
Beyond just providing more processors, the Octeon II runs faster than its predecessors -- that is, each individual core is faster than on previous Octeons. The new chip will also include more hardware acceleration blocks -- circuitry that's used for functions like deep packet inspection.
Octeon II also includes virtualization, that golden buzzword of the modern data center. In this case, it means the chip can juggle multiple applications, splitting the work among cores as necessary.
"As the traffic ebbs and flows, the applications resize themselves," Khemani says. "It also makes sure the applications don't step on each other" by vying to control the same chunks of hardware inside the chip.
— Craig Matsumoto, West Coast Editor, Light Reading
You May Also Like
Rethinking AIOPs — It's All About the DataMar 12, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Fiddling with Fixed WirelessMar 21, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Cable and 5G: The Odd Couple?Apr 18, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Delivering the DAA DifferenceMay 16, 2024