Intel makes flow processor with ARM inside

6/6/2012 3:49 PM EDT

Intel research in manufacturing process is reaping the benefits.

6/6/2012 4:07 PM EDT

This is a great processor design for networking applications, till date it was hard to think a real deep packet inspection at heavy loads on the firewall but with the development of this processor it will be possible to implement multiple filters containing deep packet inspection on the mission critical firewalls.

6/6/2012 5:13 PM EDT

I wonder if such a chip could be useful for data mining and perhaps some other search-oriented applications much like GPUs are being used for some computation-intense numerical applications.

The network processor might take incoming data, filter out uninteresting items, and route items of specific interest to specific general-purpose nodes.

I am extremely ignorant about data mining, but it seems that the decisions for network filtering, prioritization, and routing might be similar.

6/6/2012 6:09 PM EDT

Intel is simply winning by a deeper pocket which has enabled it to invest in process innovation. If Intel gets into the arms of ARM, other companies that depend on ARM better watch.

6/6/2012 7:40 PM EDT

At present this is still a matter of foundry production of SoCs that happen to include an ARM processor in a tiny portion of the die.

But it does beg the question why would I get my ARM-based SoC made by TSMC on 28-nm, if I can get it made by Intel on 22-nm?

Qualcomm may be asking themselves that question.

6/7/2012 1:14 PM EDT

The NFP-6XXX includes a legacy ARM11 multiprocessing core. I wonder if Intel is working with Netronome on an Atom core design win? http://download.intel.com/design/intarch/papers/323101.pdf

6/8/2012 7:56 AM EDT

@ip2design

Thanks IP2. You may be right

Intel did have an initiative to license Atom out and make it available through TSMC for inclusion in SoCs. That was back in March 2009

http://eetimes.com/electronics-news/4081510/Intel-to-port-Atom-cores-to-TSMC-s-tech-platform

Although we reported it was put on hold in 2010.

http://www.eetimes.com/electronics-news/4087926/Update-Intel-TSMC-Atom-partnership-on-hold

I am not sure whether there is anything fundamental that would stop Intel from letting Netronome use an Intel processor core, especially if Intel is the foundry manufacturer.

In addition it would be less royalty for rival ARM and a marketing win for Intel -- rolling back the ARM tide etc.

6/8/2012 8:01 AM EDT

@IP2design

In addition, the document referenced above by Namy44 shows there is a specific program in which Intel seeking to ease the path for the replacement of ARM within SoCs with Atom.

6/8/2012 9:49 AM EDT

It would be "a marketing win for Intel -- rolling back the ARM tide etc."

If Intel Marketing & ® Atom™ design teams are on the ball, Intel would showcase their 22nm Atom's architectural approaches "for free" with a "benchmark superior" drop in replacement for NFP-6XXX that uses Atom in place of the current ARM11 multiprocessing core.

Intel knows, big things happen one step at a time!

6/8/2012 8:54 PM EDT

umm ARM 11 is not the latest guys.

its not valid to compare old ARM to new Intel.
No reason other foundaries can't catch up quickly.

using ARM technolgy to make improvements and then claiming intel supperiority along with a die shrink advantage is ludicrous and stupid.

this discussion board is not being read by mindless marketeers.

6/11/2012 8:52 AM EDT

I asked Netronome about any plans to switch out the ARM11 processor core in a favor of an Atom core and this is what the company said in an email comment attributed to Jarrod Siket, senior vice president of marketing:

"The recently announced NFP-6xxx is primarily comprised of a large array of Netronome flow processing and packet processing cores that are optimized for network, security and content processing. The small Arm core on the device plays a limited role and is used for chip initialization."

"A unique aspect of Netronome’s flow processing architecture is that it is designed to be tightly coupled with IA/x86 processing in a heterogeneous multicore design. Our goal is to tightly couple our packet and flow processing cores with the best-in-class general purpose Intel cores via PCIe gen3, in the applications we target with the NFP-6xxx."

I interpret that as neither "yes", nor "no" nor "irrelevant."

6/11/2012 1:23 PM EDT

My take is On-chip, the ARM11 core functions are limited, as Jarrod said. The "meat and potato" of the chip is it's programmable cores, packet and flow processors. Off-chip, the NFP-6XXX programmable cores are "tightly coupled" ... "with the best-in-class general purpose Intel cores via PCIe gen3".

So, ARM11 has a cameo role on-chip role and NFP-6XXX performance comes from the tight IA/x86 based coupling between the on-chip programmable cores and off-chip Intel's BiC cores.

Someday, maybe they have a one-chip solution.

6/13/2012 5:57 AM EDT

Having developed for the existing NFP3240 I am surprised to see the ARM still in there.

Namy44 has a point in that our application and IP resides in the cores for maximum performance. We dont even bother booting the ARM. I guess some apps would use the onchip ARM for core management, slow path handoff or stacks.