Why Do We Need 800G?

In 2020 as the first widescale deployments of 400G are taking place in data centers, the industry is already thinking of 800G networking and beyond. Three technologies are being proposed to support 800G interfaces.

One of these is a pluggable option, much like 400G transceivers and the transceivers that preceded 400G.  The other two are more radical than 800G pluggable.  They promise to take high-speed interfaces to the multi-Terabit level.  They will achieve this by moving the ASIC and the optics closer together, reducing power consumption, the major factor currently constraining high data rate networking. One approach is “on-board optics,” which migrates the optics from the pluggable transceiver to a board which is also home to the ASIC.  The other approach is co-packaged optics, in which the optics and the chip share a package.

Nevertheless, the question must be asked of all 800G technologies is whether 800G is a real business.  One positive sign is that large end users – notably Microsoft and Facebook — are at the forefront of the efforts to create 800G networking technology: essentially sponsoring 800G efforts though on-board optics and co-packaged optics MSA groups.  End users know what end users need.  The goal of CIR’s new report is to examine whether 800G is a major new opportunity for the telecom/datacom industries or just a niche business at best.  In our report we take a bullish view and there are several reasons for our optimism.

Beyond the Video Boom

There are new applications emerging that will drive the need for bandwidth in the data center.  In the past, the need for higher speed data centers has been explained

by invoking a future age of digital video in which bandwidths will be permanently inadequate.  We think that the video era may have already happened.  For the past few years, video applications have been an especially important bandwidth driver and the COVID-19 pandemic has led to a videoconferencing boom.

Beyond the pandemic, however, we don’t know whether video will continue to drive bandwidth  and therefore a future massive need for 800G; 3D TV has dramatically failed in the market and it is possible that 8K television is headed the same direction. When we talk to insiders, we do not hear video, we hear how new applications will play the dominant role in driving the need for bandwidth.  The industry seems especially optimistic about 5G mobile traffic (which may itself be video) along with virtual reality (VR), augmented reality (AR), machine learning and AI.

The Need to Re-Architect the Data Center 

Hyperscale data centers will be one of the first users of 800G.  One obvious (and simple) use of 800G can be found in interbuilding connectivity.  Data centers are increasingly using multiple buildings that are closely adjacent and connected to each other at high data rates. Of course, this does not necessarily mean 800G.  Coherent detection is another possibility, as are various DWDM solutions.

As to other more complex applications, for a typical hyperscale data center widely deploying 200G servers 800G will become a necessity quite soon.  But much depends on how the hyperscale data center is architected and such architectures may vary a lot. For example, data centers that serve many external customers may have fewer geographical clusters.  Where internal use is more important, their data centers may be architected as a few gi

ant clusters. Such choices obviously impact a data center’s need for 800G and above.

In addition, a data center manager may have preferences that impact such a demand.  For example, one manager may be looking forward to 800G with a passion, while another will be more interested in an alternative 400G solution based on relative maturity of technologies.

Key Enablers:  Optics and Electronic

All the above is predicated on two key scalability enablers – one relating to the optics and the other to the switching electronics.

  • During 2020 we have seen the deployment of switches with the capacity to support multiple 400G ports. Switching technology at the chip level is outpacing expectations and it will not be long before we have the capacity to create equipment based on switches with 800G ports. The next generation of switching chips – due soon — will also be around 40 percent more space efficient, which will promote market penetration.
  • Meanwhile, the shift to 800G may not be as big a shift as it seems at first glance. In the past, data rate generations have moved in 10x factors – 1G, 10G, 100G.  But then we went to 4x – 400G, and now 2x – 800G.  The smaller the factor these easier it is to make move to the next generation.  The first 800G pluggable proposals – to put it crudely – look a bit like gluing together two (technologically mature) 400G interfaces, so we should get to 800G relatively soon

Summing up then, it seems that within a year or two we will have both the transceivers and the switching chips to rebuild the data center for an 800G world.  Better still we may have a need for such a world as bandwidth hungry 5G and even newer applications.

Scroll to Top