LightPeak, USB 3.0 and PCI Express 3.0 Announcements at IDF
Notes from Intel’s Development Forum (IDF), San Francisco, September 2010
Only a year ago at IDF-2009, Intel introduced consumer optical interconnect called LightPeak, offering 10Gbps bandwidth over a multimode fiber capable of reaching 100 meters and the potential to scale to 100 Gps. An enormous amount of press coverage, blogs, YouTube videos, and wild market forecasts followed, and numerous large and startup companies announced support for various components.
Much speculation followed of LightPeak being the next low-cost interconnect for HDTVs and PCs and perhaps with the potential very high unit volumes and low costs the technology could roll into the datacenter at one-tenth the price of current transceiver solutions and disrupt the transceiver industry. While LightCounting anticipated a big announcement, what we found at IDF-2010 this year was a significantly downplayed story.
Although LightPeak is not silicon photonics based, Intel’s silicon photonics group had several demonstrations and engineers in white coats showing prototypes of a 50Gbps silicon photonics interconnects running four lanes at 12.5 Gbps over multimode fiber. When asked when the prototype would be a viable product, the answer was “about 4-5 years out.”
This development is consistent with LightCounting’s analysis of the market for LightPeak published in December 2009. LightCounting believes there is clearly no need for 10Gbps speeds for consumer electronics at this time, other than to reduce the amount of cables in a PC by combining USB, HDMI, DisplayPort, VGA, SATA, and other I/O technology into one optical cable sharing the bandwidth. PC disk drives barely approach 1Gbps; even the fastest Flash drives operate at 2.5Gbps and “professional-level” uncompressed 1080p HDTV video at 135 Mbps. In other words, nothing in a PC even approaches the 10Gbps transfer speed of LightPeak.
While touting LightPeak, Intel had been delaying support for the next-generation USB 3.0 chips, and many speculated that Intel would not support USB 3.0 and offer LightPeak instead kicking off the consumer optics market. Over 2010, USB 3.0 garnered significant industry support and AMD incorporated it into its chipset. At IDF-2010, Intel announced its new PCI Express 3.0 Sandy Bridge architecture for next generation consumer PCs, which does not support USB 3.0 (operating at 5 Gbps), but does include support 6 Gbps SATA 6 in the chipset, and the company significantly downplayed 10 Gbps LightPeak.
Was IDF 2009 the “peak” for Intel’s optical efforts and has LightPeak been put on the back burner? Or perhaps Intel is going quiet and gearing up for a big launch at the January 2011 Consumer Electronics Show (CES)? Either way, don’t count out LightPeak and the consumer optics market yet! The industry will likely standardize on a few optical solutions eventually but it is likely to be a long road and filled with competitive optical approaches from HDMI, USB, DisplayPort and not to mention a new contender - HDBASE-T over CAT5 copper.
RELATED ARTICLE: HDBaseT = HDMI endgame? Time will tell
PCI Express 3.0 Introduction Opens Up Last Remaining Server I/O Bottleneck
While Sandy Bridge is the next generation architecture with PCI Express 3.0 for desktop PCs, Romley, is Intel’s Xeon (server) platform version. The introduction of PCI Express 3.0 addresses an important bottleneck in network traffic growth.
Enterprise networking traffic often starts at the server, and with new servers moving from two or four to eight to twelve cores (with 50 cores on the horizon) and 400Gbytes of DRAM, higher speed interconnects are sorely needed, especially in light of trends to use virtualization to increase server utilization from 20 to 90%. PCI Express 2.0, or the Gen2 version, is the backbone bus for Intel/AMD-based servers and has been a severe I/O speed-limiting factor for many switch/router systems currently shipping.
At IDF, Intel announced the Romley platform that interfaces the CPU to the PCI Express 3.0 bus and will be shipping in 2011 with data center servers and HPCs in the market early 2012. The four-year-old PCI Express 2.0, at eight lanes and 5 Gbps per lane, had an affective throughput of about 27 Gbps with all the overhead and 8B/10B encoding delays. The new PCI Express 3.0 offers 8 Gbps per lane and the capability to support 16 lanes.
Additionally, the new 128B/130B encoding with scrambling incurs very little overhead and offers features such as Quick Path, an express train that directly connects to the CPU and I/O bus and bypasses much of the bus arbitration delays; this doubles the effective throughput over PCI Express 2.0 and opens up the last remaining bottlenecks in servers in connection to the switch/router infrastructure. The 8Gbps transfer rate with the new encoding scheme is equivalent to a 10Gbps link using 8B/10B encoding.
While PCI Express 2.0 enabled dual 10GigE ports, the new PCI Express 3.0 will enable four-port 10GbE controllers, triple 4x10G QSFP, and even a 12x10G CXP port supporting 40/100GbE transceivers and AOCs for HPCs. This will enable Romley platforms to be used not only in servers but in switch, router, and storage platforms as well.
Removing the last system bottleneck is good news for 10G/40G/100G transceiver suppliers and 40G and 120G AOC suppliers, as well as Direct Attach cable manufacturers. This increases the need for higher speed uplinks in the rest of the switch/routing datacenter infrastructure all the way out to the metro link, supporting LightCounting projections for the market growth in our September 2010 Optical Communications Market Forecast Report and Database.
LightCounting is currently researching the technology and issues surrounding 10GBASE-T, 40/100G SR, and the copper–versus-optics debate in the data center in more detail in upcoming focus reports.
MORE INFO: www.lightcounting.com