Our Reports

LightCounting LLC

Our analysis is based on confidential sales data provided by leading suppliers and offers a unique port-based view of the industry.

Our industry reputation was built by providing solid market data and objective analysis to help industry executives in making tactical and strategic business decisions and see past all the market hype, rumors, press, blogs and other distortions that so often complicate and confuse many decision making issues.

Headquarters:
858 West Park Street
Eugene, OR 97401
Tel: 408.962.4851
Email: info@lightcounting.com

Global Partners

 

Our Newsletter

Sign up now to receive LightTrends, Lightcountings's Monthly Transceiver Market Updates

10GBASE-T to Dominate Gigabit Ethernet...But When?

January 2011 - $3,500

Details

purchasenow

Abstract

10GBASE-T (or IEEE 802.3an-2006) is the copper PHY (or Physical Layer) interconnect, which transforms logical data packets into raw bits that can be transmitted onto a wire, for 10-Gigabit Ethernet (10GbE) that the market has been waiting to mature and find the right power, interoperability, and latency targets to reach high volume shipments of 10GbE into the marketplace.  The standard was finally ratified in 2006, but the rollout has been spotty.  Now that the 40nm versions of 10GBASE-T PHYs are sampling, LightCounting believes that 10GBASE-T will begin to drive major growth in 10GbE ports in servers and switches, but not nearly as quickly as some believe.  Because we believe that integrated controller/10GBASE-T PHY chips will not be ready for the next Intel-based server platform design cycle (code named Romley) that will ship in the latter half of 2011, LightCounting’s forecast for 10GBASE-T does not expect 10GbE shipments to surpass 1GbE shipments until 2014.

Executive Summary

10GbE has been waiting a long time to transition from being the heir apparent to a ubiquitous replacement for 1GbE.  This report provides a forecast for 10GBASE-T—for adapters and LOMs (LAN on Motherboard), and 10GBASE-KR (for backplanes used in blade servers).  Here are some of the many reasons behind this forecast. 

Optical 10GbE port shipments have finally started to take off:  Going forward, SFP+ will be the dominant optical interface.  10GbE SFP+ port shipments with a reach of less than 300 meters that are used in server to switch and some switch-to-switch applications have tripled over the past three quarters, and LightCounting expects the high growth trend to continue. 

10GBASE-T for adapters becomes feasible in late 2011:  Because the use of 10GBASE-T PHYs, on dual port server adapters (the predominant version of shipping adapters), has not been feasible to date and optical interconnects are much more expensive than copper once these products reach the end user, the market for 10GbE has been stunted.  With the newest versions of 10GBASE-T PHYs, developed using the 40nm semiconductor process, the power constraints of 10GBASE-T are finally being overcome, enabling dual-port 10GbE adapter sales to begin in earnest.  However, the mass rollout of 10GbE interconnects will not occur until most rack servers are equipped with 10GbE LOMs begin shipping in 2014.


10BASE-KR for Blade Servers has been shipping since 2009:  In March 2009, HP announced the first 10GbE LOM in its Blade servers under the BladeSystem brand when it refreshed its blade servers with Intel’s new Nehalem family of processors.  HP used Broadcom 10GbE controller chips with 10GBASE-KR PHYs.  Once Blade servers offered LOM implementations of 10GbE, 10GbE port volumes grew dramatically.  The significance of LOM is huge.  Previous to this generation of blade servers, 10GbE could only be enabled via adapter cards. 


Integrated controller with dual 10GBASE-T PHYs too late for Romley cycle in 2011:  At the Intel Developer Forum (IDF) in September 2010, Intel spent a large portion of the forum showcasing its upcoming Sandy Bridge Microarchitecture due to ship for desktops and mobile platforms in the first half of 2011 and for servers in the second half of 2011.  The server platform is called Romley.  Because server platforms take nine months or more to validate (i.e., to test the server, switches, internal and external storage, and peripheral hardware with all major software platforms such as those provided by Microsoft, Linux distributions, VMware, and Oracle), the design win opportunity for a chip company to design its chips into a given OEM’s server will close in early 2011.  Further, the server OEM must have the final, shippable version of all semiconductor components available to the OEM before validation can begin.  Although Intel did demonstrate its integrated 10GbE controller with two 10GBASE-T PHYs at IDF, we at LightCounting do not believe that the chip will be designed into LOM implementations beyond a few server models by major server vendors for shipment in 2011.  We also do not expect any other competitors such as Emulex, QLogic, or Broadcom to have a working integrated part ready before spring of 2011.  As a result, we believe that the opportunity for 10GbE LOM across the major server OEMs (i.e., HP, IBM, or Dell), Romley servers is small.  Some minor server OEMs may put Intel’s LOM on some of their server models, and one or more of the major OEMs may test the market with a server model or two.  However, we believe that the next opportunity for a major wave of 10GBASE-T LOM on Intel-based servers is in early 2013 for design wins and in early 2014 for shipments coincident with the Haswell microarchitecture.


What do the large OEMs want?  Answer- profits!  All x86 servers ship with at least one CPU from either Intel or AMD, at least a few gigabytes of memory, and in almost all cases, at least one disk drive.  It is hard to differentiate one OEM’s Intel or AMD CPU or memory or disk drive; consequently, the I/O (input and output) peripherals and architecture is where the money is.  Over the past ten years, Fibre Channel became a very profitable part of the server OEM portfolio.  OEMs do not want to lose the opportunity to upsell I/O hardware and software solutions by quickly commoditizing the I/O by putting 10GbE as the standard (free) I/O solution.     

   
Would server OEMs rather sell 10GbE adapters or LOMs?  Adapters.  OEMs realize high profit margins on I/O solutions such as Fibre Channel, InfiniBand, and 10GbE adapters and the associated management software.  Inevitably, they will be forced to provide 10GbE as a LOM as weaker competitors begin to offer it to gain share from the large OEMs, but they are in no hurry to do so. 


Would OEMs rather sell optical or copper?  Optical (the markup on transceivers is huge).  Final end user price is another huge factor that will drive 10GBASE-T adoption.   Typically, server, switch, and storage OEMs purchase SFP+ optical transceivers in volume for $50–60 today.  However, that is not close to the price that end users pay.  Because the OEM ties the use of their branded transceivers with their products via either maintenance contracts or firmware that checks the vendor data burned into the transceiver, the user typically must buy transceivers from the server, switch, or storage vendor.  This allows the OEM to mark up the cost of the transceiver astronomically.  Prices on the web range from $120 or so for unbranded SR modules, which the OEM typically will not support, to $500 and sometimes $1,200 or more for OEM-branded SR modules.  For the end user, the beauty of 10GBASE-T is that no one can mark up the PHY, because it is the standard I/O connection preinstalled on the server (as 1GbE is today) rather than an add-in component that the OEM can charge for. The result is a huge disparity in port pricing for optical versus copper ports.


Are converged fabrics or virtualization driving demand for 10GbE?  Virtualization.  Virtualization is a new phenomenon in the implementation of servers over the past five years driven primarily by VMware.  Before virtualization, most x86 servers ran a single instance of an OS and a single application.  When Intel could not increase its processor clock speeds anymore in the 2000’s due to the ever-rising power needs of its single core processors, the company started to add multiple cores as its budget for more transistors increased.  With VMware’s ESX VM Hypervisor in place, a user can run many workloads on a single machine that used to be run by many physical servers.  Microsoft, Citrix, Red Hat, and SUSE also offer VM Hypervisor technology in their products.  Servers based on Intel’s newest processors can replace as many as 15–20 older machines, and the savings on power and cooling for the user is huge.  In fact, server OEM executives tell us that the primary reason for sales of new x86 servers is server consolidation enabled by virtual machines.  Further, the server OEM has an opportunity to sell not only dual Ethernet ports, but dual Fibre Channel or InfiniBand ports as well, and the associated switches.


10GbE shipments will not exceed 1GbE shipments until 2014 when 10GBASE-T explodes:  When the Haswell generation of servers ships in 2014, total copper 10GbE ports will grow from almost 17 million in 2013 to over 37 million in 2014.  Total 10GBASE-T ports will explode in this same timeframe from just over 6 million in 2013 to over 24 million in 2014.  10GBASE-T adapter ports will finally become a strong market in 2011 and beyond, with a compound annual growth rate of 233% versus optical adapter port growth of 55% from 2010 through 2014.   Further, copper 10GBASE-KR port shipments on Blade servers enable copper 10GbE to surpass optical 10GbE ports in 2010. 


10GBASE-T and 10GBASE-KR Revenues to OEMs Grow Dramatically from 2009 to 2014: 
10GBASE-T chip developers will finally achieve strong sales from 2011 through 2014 culminating in over $350 million in chip sales to OEMs in 2014.  The compound annual growth rate of 10GBASE-T revenues will be 220%.  10GBASE-KR chip makers see strong growth early in the forecast period, but during 2011, the market shifts to an embedded PHY in the 10GbE controller chip with revenues derived from the PHY dropping dramatically.  The same phenomenon also occurs for 10GBASE-T in 2014, but because the volumes grow so dramatically in 2014, revenues also grow from $132 million in 2013 to $351 million in 2014, a growth rate of 165%.

Table of Contents

  • Abstract
  • Executive Summary
  • Chapter 1: 10 GBASE-T market Finally Ready to be a Factor in 10GbE Adapters
    • History and Development of 10GBASE-T
    • End user benefits of 10GBASE-T
    • Markup on optical transceivers is huge!
  • Chapter 2: 10GBASE-KR Already a Factor in Blade Servers
    • Taxonomy of servers
    • 10GBASE-KR LOM for blade servers has been shipping since 2009
    • Blade server backplane architecture allows first implementation of 10 GbE LOM
  • Chapter 3: Semiconductor Timeline for 10GBASE-T and for High Volume Servers
    • Newest 10GBASE-T silicon (40nm) finally meets PCIe adapter power needs
    • 10GBASE-T LOM will change everything, eventually
    • Supplier Design Win Window: It's all about Intel's server platform timeline
    • Integrated controller with dual 10GBASE-T PHYs too late for Romley cycle in 2011
  • Chapter 4: What do the Large OEMs Want?
    • Would server OEMs rather sell 10 GbE adpters or LOMs
    • Are converged fabrics or virtualization driving demand for 10 GbE?
  • Chapter 5: 10GBASE-T Competitive Landscape
    • Applied Micro (AMCC)
    • Aquantia
    • Broadcom
    • Marvell
    • Solarflare
    • PLX/Teranetics
  • Chapter 6: Forecast and Analysis
    • The 10GBASE-T analysis and forecast methodology
    • Market erosion and overall expansion
    • Demand factors
    • 10GBASE-T market forces and forecast unknowns
    • Market-driving forces
    • Market limiters
    • Unit shipments forecast and analysis
      • 10GBASE-T to grow dramatically in 2014
      • 10 GbE Phys to change dramatically from 2011 to 2014
      • 10 GbE surpasses 1 GbE in server port count in 2014
      • 10GBASE-T and 10GBASE-KR Revenues to OEMs grow dramatically from 2009 to 2014
  • Conclusions

Figures

  • Figure 1: 10 GbE (<300M) Transceiver Shipments by MSA
  • Figure 2: SFP+ Adaper, module and Direct Attach Cable
  • Figure 3: The Blade Server System Backplane
  • Figure 4: Approximate Unit Distribution of Types of x86 Servers
  • Figure 5: Forecasted Port (Server and Switch/Router) Volumes of 10GBASE-KR
  • Figure 6: 10 GbE Optical versus 10GBASE-T Server Adapter Ports
  • Figure 7:Intel's Tick-Tock Model for Server Platforms
  • Figure 8: Typical Virtual Machine Architecture
  • Figure 9: IBM's Virtual Fabric Implementation of vNIC
  • Figure 10: Unit Shipment Forecast for 10 GbE Adapters and LOMs
  • Figure 11: Unit Shipment Forecast for 10 GbE Adapters and LOMs
  • Figure 12: 10GBASE-T versus 10GBASE-KR
  • Figure 13: Server Ports of 10 GbE versus 1 GbE
  • Figure 14: 10 GBASE-T and 10 GBASE-KR Revenue to OEMs

Tables

  • Table 1: Key Data Center Networking PHYs Going Forward
  • Table 2: 10GBASE-T Adapters Cards: Total Watts
  • Table 3: 10 GbE Modes

Companies Mentioned

  • Applied Micro (AMCC)
  • Aquantia
  • Broadcom
  • Marvell
  • Solarflare
  • PLX/Teranetics