& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
Page 1 / 2   >   >>
sgan201
sgan201
12/5/2012 | 2:28:24 AM
re: Internet Machines Starts Anew
Hi,
Success is judged by revenue and number of units sold.. Intel NPU are used by Cisco, Linksys, and Netgear. That fact alone means the quantity will be higher than MSPP or DSLAM..
http://www.intel.com/design/ne...

In my book, off-the-shelf NPU are very successful..

Dreamer
krisman
krisman
12/5/2012 | 2:28:24 AM
re: Internet Machines Starts Anew
I have a question to the wise.

Will off-the-shelf NPUs ever be succesful in mainstream Telecom and Datacom applications such as MSPPs, WAN Rtrs, and DSLAMs. Or will they be plagued by too much power and programming complexity.

Thanks

x-man
x-man
12/5/2012 | 2:28:23 AM
re: Internet Machines Starts Anew
Intel went in to Cisco and did a very good job
of selling at the executive level. They did so
good that nobody else was considered. Then when
the implementation teams tried to use the stuff
it fell flat on its face. There are many things
that you should not try to do with risc microprocessor cycles. Intel does not yet
understand this.

Agere is actually the market leader in off-the-shelf net processors. Their stuff has basically
the same problem.

Bottom line: you cannot get enough performance doing everything with microcode. You cannot get
enough flexibility to satisfy a large customer
base with pure hardware solutions. So companies
like Cisco and others will continue to have their
home grown efforts in order to get both the
performance and features that they need.

>Hi,
>Success is judged by revenue and number of units >sold.. Intel NPU are used by Cisco, Linksys, and >Netgear. That fact alone means the quantity will >be higher than MSPP or DSLAM..
>http://www.intel.com/design/ne...

>In my book, off-the-shelf NPU are very >successful..

>Dreamer
sigint
sigint
12/5/2012 | 2:28:22 AM
re: Internet Machines Starts Anew
x-man:
Intel went in to Cisco and did a very good job
of selling at the executive level. They did so
good that nobody else was considered. Then when
the implementation teams tried to use the stuff
it fell flat on its face. There are many things
that you should not try to do with risc microprocessor cycles. Intel does not yet
understand this.
_________________________________________________
and, you might add, they stepped back, sent their engineering teams to interact directly with the Cisco folks - whose inputs were used to reshape the device significantly. The interaction amongst engineers worked to everyone's advantage. The device is in a much better shape now.
Tony Li
Tony Li
12/5/2012 | 2:28:22 AM
re: Internet Machines Starts Anew
Bottom line: you cannot get enough performance doing everything with microcode. You cannot get enough flexibility to satisfy a large customer base with pure hardware solutions.

-----------------------


Ok, I disagree violently.

You can get plenty of performance with microcode IF you designed the hardware to give you sufficient performance. I agree that the current off-the-shelf NPUs have insufficient performance to address their intended markets. However, we have a proof of existance that it is possible to construct NPUs with sufficient performance that are ucoded. Such an NPU is a serious strategic advantage.

Tony
sgan201
sgan201
12/5/2012 | 2:28:21 AM
re: Internet Machines Starts Anew
Hi Folks,
It is a volume game.. The largest number of router sold right now is the cheap Linksys router. So, if you have design wins in Linksys and Netgear, you will sell more NPUs to routers than anyone else..
Performance is relative.. If your target is high volume market like broadband router, a RISC core works perfectly fine..

Agere Network Processor is too high end.. The market for high end network processor is so small that they cannot sell enough to make money..


Dreamer
optoslob
optoslob
12/5/2012 | 2:28:19 AM
re: Internet Machines Starts Anew

Tony,
I agree that ucoded NPU's can be made to scale to whatever load, IF the system is designed properly. HOWEVER a ucoded application is typically 3 to 5 times the power of dedicated hardware.
Ti DSP's learned this lesson the hard way, Ti internally had a policy of using DSP's for all possible signal processing applications, trouble is people like Broadcom with dedicated signal processing functions were kicking their butt. These days no one would dream of doing a Gige-phy or WLAN Phy without dedicated decoders. The trick is to maintain the flexibility of a DSP while meeting the power and cost constraints of the application.

optoslob
lightbulb0
lightbulb0
12/5/2012 | 2:28:12 AM
re: Internet Machines Starts Anew
My take on this is, it is not a technical issue at all. It is all about market timing. NPU was ill-fated at the very beginning. The networking business is not like general computing industry. They done have the elasticity and scalability. You must get 1M/100M/1G/10G/100G. There is nothing between. The application nature itself demand cost-effective and stringent solution. NPU has zero chance becoming mainstream.
PO
PO
12/5/2012 | 2:28:05 AM
re: Internet Machines Starts Anew
All NPUs are created equal. And they'll never be mainstream. And I think I saw some monkeys flying behind me.

NPUs are great for what they're designed for. But therein lies the rub: different vendors' NPUs are better for different applications. And very few people in the world understand them well enough to program them correctly, let alone select the right one for their particular purpose.

And yes, some functions are best done in dedicated hardware. Others require the flexibility of a programmable device. Still others are somewhere in between, and want an FPGA-based solution. There are no simple answers; rather, a design needs to consider the maturity of the problem space, the variety of the options involved (i.e. user configurability), processing and I/O loading, queueing strategies employed (and overload management), time to market, and cost issues.

Oh for the days when there was only one answer!
Outsider
Outsider
12/5/2012 | 2:28:05 AM
re: Internet Machines Starts Anew
If IMC can do it then why can't all failed Start-up's simply start anew with a fresh product line?

Why couldn't Silicon Access venture off into this PCI Express market? They were a much technically deeper company than IMC.

By the way, what was the point of replacing Chris Hoogenboom with Frank Knuettel, who has absolutely no experience in this sector at all.

Outsider
Page 1 / 2   >   >>


Featured Video
Upcoming Live Events
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events