June 7, 2006
Intel to sell off it's Communication chip business, one I got wrong
As a venture capitalist you are basically making calculated bets. One of my bets was development of an operating system for NPU communication chips which Intel had made a big commitment to. It is becomming clear this week: MercuryNews.com | 06/03/2006 | How Intel wasted billions that Intel is getting out of the business after spending over $10B on it. The start-up I funded four years ago, IPFabrics, was a very good bet at the time to draft behind this investment. But one of the major risks was always that Intel would not be sucessful and would get out of the business. That has come to be. That doesn't mean that the bet on IPFabrics was wrong. In fact, they have created some very valuable technology which has the potential to be repurposed in a different way. That is also a fact of life in venture investing. The business plan that ends up making money may not be the one you originally invest in. Gosh this is fun.
Posted by Martin at 10:33 AM | Comments (0) | TrackBack
September 7, 2005
Intel committed to the NPU market
While press around the NPU market is sparse, here is a recent round-up article from light reading: Light Reading - Networking the Telecom Industry. Basicallly the market has been access products. The largest (volume) seller in Intel's line is the Westport which is more of an edge or access speed processor. The core router stuff (the OC192 market) is dog slow. The main problem with the chips continues to be "programmability" which, thankfully IP Fabrics has partially solved. Hopefully as more people understand that the progammability problem is solvable, they will implement these programmable chips deep into the network. It needs to happen. We need a standard PC economics programmable hardware platform as far into the core of the network as possible.
Posted by Martin at 12:15 PM | Comments (0) | TrackBack
June 9, 2005
Apple's move to Intel affects Freescale and maybe NPUs
As with most of the industry, I was quite surprised to hear that Apple was going to dump Power PC and move to Intel. It must have been a hard thing for their ego to swallow, but in the end the superior economics of the X86 platform were just too compelling. Apple is more clearly turning into an OS company now. Out with the proprietary hardware. Apple OSX is basically Unix with a better UI. So would you buy Windows or Apple on an Intel platform? Hummm.
An interesting thing to note is what the loss of Apple as the last major customer of the Power PC platform will do to further investments on that chip line. It can't be good. Here is Linley's take.
===Apple Decision Impacts Freescale, IBM
-------------------------------------------
Apple's decision this week to move its Macintosh computers from PowerPC to
x86 in 2006 will impact Freescale and IBM to a modest extent, but it will not do significant harm to their embedded-processor strategies. We estimate Freescale's revenue from Apple was about $150 million in 2004, but the impact goes beyond this revenue; without Apple as a customer, the company is unlikely to continue developing new products in its MPC744x line of standalone processors. We believe this product line generates another $50 million or more in annual revenue from customers other than Apple.
Fortunately, Freescale no longer uses the CPUs from its 744x line in its most of its popular PowerQuicc products. The e600 (G4) CPU from the 744x dissipates too much power for most embedded applications, so Freescale developed the e500, which delivers slightly less performance but uses much less power, for PowerQuicc. Without Apples funding, we expect Freescale to focus on processors that provide a balance between performance and power dissipation rather than very high-performance products. We also expect Freescale to redirect development of the e700, a 64-bit PowerPC CPU intended to compete with IBMs G5 CPU, to better meet the demands of the embedded market, which is likely to further delay the project.
With Freescale unable to deliver a G5-class product in 2005, Apple has been forced to rely on IBM, but that company has become less and less interested in developing standard products. Over the past two years, IBM has sold its network processor, switch fabric, TCAM, and integrated PowerPC products, leaving only the standalone processors that it sells to Apple and a few others. But apparently even the estimated $250 million that IBM received from Apple in 2004 was not enough to keep Big Blue's attention. Apple's decision should allow IBM to drop the final vestiges of its standard-product business, letting it focus entirely on video-game manufacturers and its other large ASIC customers. By the way, anyone who believes IBM will be selling Cell processors to hundreds of customers in a few years just isn't paying attention. --LG
Posted by Martin at 1:54 PM | Comments (0) | TrackBack
May 22, 2005
More meat for the NPU fire
Haven't posted about NPUs for some time. I remain extremely bullish on my investment in IP Fabrics. Last week Raza Microelectronics finally "officially" unveiled their high end NPU which I had heard rumors for some time. No surprise that Warburg Pincus is behind it with Beau Vrolyk on the board. Beau an I have often bashed brains about the future of NPUs and what will happen with the hardware and software. Nice to be able to talk publicly about something I have known about for over a year. This is a good thing for the NPU market. After a year of major vendor pull-back from the NPU market and Intel taking all the marbles, it is good to have a new entrant.
Here is Linley's summary of the announcement:
===Raza Discloses Powerful Processor, EoS Chip
--------------------------------------------------
At Spring Processor Forum, stealth startup Raza Microelectronics (RMI) announced its much-anticipated multicore processor, known as the XLR. This powerful device contains eight MIPS64 processors operating at speeds up to 1.5GHz. Although it was announced only this week, the XLR has been sampling since December.
Like most network processors, the XLR uses multithreading to improve the efficiency of its scalar CPUs while processing many packets in parallel.
(David Hass, the XLR's architect, was formerly an architect at Nexsi, a failed NPU startup also funded by Raza.) Each XLR CPU switches among four threads, for a total of 32 threads on the chip. The XLR is the only commercially available MIPS processor to implement multithreading.
Along with its high-speed CPUs, the XLR contains 2MB of cache memory and a crypto engine capable of 10Gbps of AES, 3DES, SHA-1, or MD5 encryption. It supports 12.8GB/s of peak bandwidth to DDR2 SDRAM or RLDRAM. The processor connects to the rest of the system through two integrated 10GbE MACs or four integrated GbE MACs as well as other high-speed interfaces. RMI quotes a list price of $850 for the high-end XLR but offers lower-cost versions with fewer CPUs and other restricted capabilities.
The company positions the XLR for a wide range of control-plane and data-plane applications. The chip is well suited for high-end control-plane designs; compared with Broadcom's BCM1480, the XLR offers more CPUs, more clock speed, and multithreading for a lower price. We do not, however, expect the XLR to displace traditional NPUs from Layer 3/4 routing applications, due to the XLR's greater cost and power dissipation and incomplete software solution. Instead, we see the XLR competing with Cavium's Octeon in data-plane applications that require greater per-packet processing, such as security (intrusion prevention/antivirus), IP storage, and web switching. Octeon lacks the XLR's multithreading capability but offers a hard-wired DFA engine, useful in many security applications.
At the same time, RMI also introduced its Orion chip, designed to provide packet service over the Sonet/SDH network. The Orion family includes products for OC-3, OC-12, or OC-48 Sonet rates and 8 FE ports, 2 GbE ports, or 3 T3/E3 packet ports. The Orion architecture consists of a Sonet/SDH mapper and an Ethernet interface. The mapper includes a Sonet/SDH framer, pointer processor, VCAT (virtual concatenation), LCAS, and GFP framer. The Ethernet interface includes MACs, MAC forwarding, and a traffic manager for 2K flows. The chip may be used for Ethernet over Sonet (EoS) or to switch Ethernet traffic.
Although vendors such as PMC-Sierra and TranSwitch have been shipping data-mapping products for similar applications, RMI offers a highly integrated chip with sophisticated traffic-management capabilities that include dual-bucket policing, random early discard, and shaping. As a late entrant, RMI's challenge is to use Orion's unique features to establish a position in this nascent market. --LG/JB
Complete coverage of Cavium's Octeon and similar processors appears in our report "A Guide to High-Speed Embedded Processors."
http://www.linleygroup.com/Reports/ctlpln_guide.html
Posted by Martin at 8:24 PM | Comments (0) | TrackBack
February 23, 2005
Doug Comer on NPUs...
Doug Comer, super champion of NPUs, on leave from Purdue to Cisco, has written a very comprehensive overview of the NPU market, how and why you need an NPU, and where we are going. Network Processors - Volume 7, Issue 4, December 2004-The Internet Protocol Journal - Cisco Systems. Doug also has some nice words for IP Fabrics.
"Another pleasant exception comes from IP Fabrics, which has focused on building tools to simplify programming. Like Agere, IP Fabrics has developed a high-level language that allows a programmer to specify packet classification and the subsequent actions to be taken. The language from IP Fabrics is even more compact than the language from Agere."
thanks doug!
Posted by Martin at 2:03 PM | Comments (0) | TrackBack
April 23, 2004
NPU market grows 30% in 03
The Linley Group is reporting that the NPU market grew 30% in 2003, the first growth since 2000. AMCC still leads the pack with 40% share. They just bought IBM's PowerPC npu business, so their share will grow. Intel doubled their share after shipping the IXP2400 in Q303. But all told the silicon business was only $85M last year. Still a puppy. Look for that to grow about 50% in 04 says Linley. And right in the middle of this year, my company, IP Fabrics will have their alpha virtual machine out. Looks like good timing!
Posted by Martin at 10:05 PM | Comments (1) | TrackBack