SAN JOSE, Calif. – Broadcom Corp. leveraged its NetLogic acquisition to roll out what’s being seen as the highest performance communications processor on the market. The XLP980 leapfrogs archrival Cavium Networks to become the first to handle a trillion network operations/second.
The 28nm chip uses 20 quad-threaded, out-of-order MIPS cores to drive throughput to 160 Gbits/s, scalable to 1.28Tbits/s using multi-chip coherency. It provides hardware support for virtualization, security, deep-packet inspection and quality-of-service features.
The XLP980 also is the first multicore processor to integrate 40G Ethernet ports, said Bob Wheeler, a senior analyst with The Linley Group (Mountain View, Calif.). The comms chips compete with ASICs from big systems companies such as Cisco Systems and increasingly with Intel’s Xeon processors.
“Broadcom has grabbed the performance lead [because] Cavium has not yet taped out the CN78xx high-end Octeon III," Wheeler said. The XLP980 is “a beast, however, so it will serve only applications that can afford the high price and power dissipation such as EPC gateways, high-end security appliances and blades for service-provider routers,” he said.
Rising data consumption on wired and wireless networks is driving demand for such high-end chips said Wheeler at the Linley Carrier Conference where Broadcom announced its chip.
Mobile traffic grew 70 percent in 2012, making up 15 percent of all Internet traffic today, Wheeler said. Mobile traffic will more than double to 30 percent of Net traffic by 2015, he said.
The bad news is bandwidth is increasing faster than carrier revenue. Thus service providers are putting pressure on systems makers to reduce cost per Mbyte, a factor that could drive more OEMs to turn to merchant chips.
In wired networks, broadband subscribers are projected to rise from 642 million in 2012 to 940 million in 2018. DSL handles 63 per cent of the traffic now but it is on a gradual decline.
Various flavors of fibre networks are on the rise with passive optical nets expected to grab the lion’s share in the long run. China already has more wired broadband subscribers than the U.S. and is projected to have three times more subs than the U.S.
The XLP980 and the CaviumCNxx are similar in so many ways but yet different. They both seem to have a slew of accelerators specific to the networking segment, both use MIPs based CPUs and are on 28nm. But the CPU implementations are different. Netlogic's is an out-of-order multi-thread while Cavium is an in-order single-thread. I wonder which is better suited for the target application.
Interestingly the Intel processors are also out-of-order multi-threaded so it might appear that the XLP can better compete with Intel.
Intel, Comcast, Cisco just announced the XFINITY Wireless Gateway. It uses Intel's Puma 6 silicon. The Wireless Gateway from XFINITY Internet is capable of up to 16 bonded downstream channels supporting speeds up to 640 Mbps.
The Puma 6 is the first DOCSIS silicon to reach up 1 Gbps downstream by bonding up to 24 DOCSIS channels and up to 240 Mbps upstream by bonding up to 8 DOCSIS return channels
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.